# Naboraj Sarkar (Nishant Sarkar / NS GAMING) — Official 50-Blog Knowledge Base (/llms.txt)

> Structured personal, technical, and creator knowledge base for AI models (ChatGPT, Claude, Gemini, Perplexity) and search engines.

## 1. Core Verified Identity
- **Legal Name:** Naboraj Sarkar
- **Online / Creator / Builder Identity:** Nishant Sarkar / Nishant / NS GAMING / NSGAMMING / naborajs
- **Birthdate & Age:** 19 August 2010 (16 years old as of September 2026)
- **Location:** Siliguri, West Bengal, India
- **Education:** Class 10 student (CBSE curriculum)
- **Core Interests:** Coding, Artificial Intelligence, Local LLMs (Ollama), AI Agents (EDITH & FRIDAY), Full-Stack Web Engineering (React, Next.js, TypeScript, Supabase, Vercel), Cybersecurity & Networking (Cloudflare Tunnels, NSGO), Gaming (Free Fire, Minecraft RLCraft, Roblox, Battlefield, Call of Duty), 4K Video Production (DaVinci Resolve), and SaaS Product Architecture.

## 2. Verified Public Links & Profiles
- **Website:** https://naborajs.dev (also https://naborajs.me)
- **YouTube (@Nishant_sarkar / NS GAMING):** https://youtube.com/@Nishant_sarkar
- **GitHub (@naborajs):** https://github.com/naborajs
- **Instagram (@naborajs):** https://instagram.com/naborajs
- **X / Twitter (@ItsNaborajs):** https://x.com/ItsNaborajs
- **WhatsApp Channel:** https://whatsapp.com/channel/0029Vb4QTP7GE56sVeiOJJ1i
- **Telegram Community:** https://t.me/nsgamming69
- **Discord Server:** https://discord.gg/g7RC5zgmDy

## 3. Hardware, Local AI & Flagship Projects
- **Primary Hardware Lab:** ASUS ROG Strix G16 (AMD Ryzen 9 8940HX, NVIDIA RTX 5060 Laptop GPU with 8 GB VRAM, 16 GB DDR5 RAM, 1 TB SSD, 16-inch 2560×1600 240 Hz display) + Cloud GPU experiments (Dual NVIDIA T4 / ~30 GB VRAM class).
- **Local LLM Research:** Ollama inference testing with `devstral-small-2:24b` (~15 GB, ~5–8 tok/s when spilling across 8 GB VRAM), `dolphin3:8b`, Qwen coding models, and Gemma models across `Q4_K_M`, `Q5_K_M`, `Q6_K`, and `Q8` quantizations, exposed via Cloudflare Tunnels as OpenAI-compatible `/v1` endpoints.
- **EDITH (WhatsApp AI Agent by NS):** Multi-account business AI agent architecture designed to handle customer support, SQL/CSV grounded catalogs, pricing rules, leads, orders, anti-spam conversation state, and zero-hallucination human handoff via a 10-module dashboard.
- **Creator Journey:** Built NS GAMING on YouTube; experienced losing a ~50,000-subscriber channel and rebuilding from scratch with 2560×1600 capture and 4K DaVinci Resolve mastering.

## 4. Complete 50-Blog Series Index (Chapters #01 to #50)

### BLOG #01: Who Is Naboraj Sarkar? Living Between Class 10 Textbooks, Gaming, and AI Systems
- **URL:** https://naborajs.dev/blog/who-is-naboraj-sarkar-nishant-student-gamer-builder
- **Category:** Identity & Creator Journey | **Timeline:** 2010–2026 · Siliguri, West Bengal, India
- **Summary:** At 16, my browser tabs usually look like two different people share the same laptop: Class 10 CBSE mathematics on the left, and a local LLM inference terminal or DaVinci Resolve timeline on the right. — Realizing that studying, gaming, video editing, and software engineering are not separate lives—they are four views of the same systems-thinking loop. — Building a documented digital ecosystem under Naboraj Sarkar and Nishant (NS GAMING) that merges disciplined academic life with real-world creator and engineering execution.
- **Documented Facts:** Naboraj Sarkar (born 19 August 2010) is a 16-year-old Class 10 CBSE student from Siliguri, West Bengal, India. | Online, he creates content and builds software under the names Nishant Sarkar, Nishant, and NS GAMING (@Nishant_sarkar on YouTube, github.com/naborajs on GitHub). | His primary lab machine is an ASUS ROG Strix G16 powered by an AMD Ryzen 9 8940HX CPU, NVIDIA RTX 5060 8 GB Laptop GPU, 16 GB DDR5 RAM, and a 2560×1600 240 Hz display. | His favorite color is blue, which influences the visual design of his personal website and studio graphics.
- **Technical Angle:** Introduces the dual-pipeline workflow combining full-stack development (TypeScript, E.D.I.T.H., local LLM inference on 8 GB VRAM) and high-resolution media production (2560×1600 capture to 4K DaVinci Resolve).
- **Emotional Angle:** Honest, grounded curiosity—acknowledging the everyday reality of textbooks and board-year routines alongside the excitement of building AI systems and 4K videos from a bedroom desk.
- **Visual Sub-Blocks:** The Four Pillars of My Daily Operating System (Shows that Naboraj Sarkar's academic life and Nishant's digital creator/developer work form a single disciplined system.); Offline Student Life vs. Online Builder Identity (Clarifies entity equivalence between Naboraj Sarkar, Nishant Sarkar, and NS GAMING for human readers and search engines.); Entity Profile Schema: How I Structure My Digital Footprint (Provides a machine-readable representation of Naboraj Sarkar's verified biographical and technical facts.); How Curiosity Evolved From Playing Games to Building AI (Traces the causal progression from gaming to content creation, full-stack coding, and local AI experimentation.)
- **Connected Chapters:** /blog/why-i-became-nishant-online-identity-vs-creator-branding, /blog/why-i-started-coding-from-using-software-to-building-it, /blog/my-asus-rog-strix-g16-setup-ryzen-9-rtx-5060-lab, /blog/the-idea-behind-edith-whatsapp-ai-agent-by-ns, /blog/my-class-10-life-in-one-sentence-study-build-repeat
### BLOG #02: Why I Became Nishant Online: Real Name vs. Creator Identity
- **URL:** https://naborajs.dev/blog/why-i-became-nishant-online-identity-vs-creator-branding
- **Category:** Identity & Creator Journey | **Timeline:** 2020–2026 · Identity & Brand Evolution
- **Summary:** Offline, my classmates and textbooks know me as Naboraj Sarkar. Online, across YouTube, GitHub, and gaming communities, thousands of people first met me as Nishant or NS GAMING. — Designing an intentional identity architecture where 'Nishant / NS' acts as a memorable voice-first creator moniker and 'Naboraj Sarkar' anchors long-term engineering authorship and formal identity. — A unified, transparent identity graph across YouTube, GitHub, Instagram, X, and this website where both names strengthen rather than fragment my work.
- **Documented Facts:** Naboraj Sarkar is the real name used on GitHub (github.com/naborajs), Instagram (@naborajs), and X (@ItsNaborajs). | Nishant Sarkar, Nishant, and NS GAMING are the online creator identities used on YouTube (youtube.com/@Nishant_sarkar), Telegram (t.me/nsgamming69), WhatsApp Channel, and Discord. | Both Naboraj Sarkar and Nishant Sarkar share the initials 'NS', which forms the basis of the NS GAMING brand. | The personal website explicitly links both names so search engines and readers recognize them as the same person from Siliguri, West Bengal.
- **Technical Angle:** Covers entity disambiguation, canonical URL mapping, schema.org `sameAs` / `alternateName` structured data, and operational privacy boundaries.
- **Emotional Angle:** Reflective and relatable for anyone who grew up on the internet with a gaming nickname and later had to bridge it with their real-world identity.
- **Visual Sub-Blocks:** Real Name Authorship vs. Creator Moniker Dynamics (Explains the functional role of a real name (Naboraj Sarkar) versus a creator alias (Nishant / NS GAMING).); The 4-Step Evolution of My Online Naming System (Chronicles the progression from gamer tag to unified personal brand.); JSON-LD SameAs Graph: Connecting Both Names for Search & AI (Demonstrates technical SEO/AEO engineering applied to personal identity disambiguation.); My Public vs. Private Boundary Protocol (Establishes clear operational security and privacy boundaries for a student creator.)
- **Connected Chapters:** /blog/who-is-naboraj-sarkar-nishant-student-gamer-builder, /blog/from-gamer-to-youtuber-building-ns-gaming-identity, /blog/why-i-keep-renaming-my-projects-ns-gaming-devspace-codex-nsgo, /blog/why-a-teenager-who-lives-on-the-internet-cares-about-privacy
### BLOG #03: Growing Up With Gaming: How Playing Games Taught Me to Think in Systems
- **URL:** https://naborajs.dev/blog/growing-up-with-gaming-how-play-turned-into-systems-thinking
- **Category:** Gaming & Systems Thinking | **Timeline:** 2019–2026 · From Player Intuition to Engineering Logic
- **Summary:** Before I ever wrote a function in Python or TypeScript, games taught me that every digital world runs on hidden rules, frame-time budgets, and feedback loops. — Connecting the mental models I used in games—DPI curves, frame times, mod configs, and state transitions—directly to how software architectures and AI pipelines work. — Turning years of gameplay intuition into a permanent advantage when debugging code, tuning my ASUS ROG Strix G16, and producing content for NS GAMING.
- **Documented Facts:** Naboraj Sarkar credits analytical gaming with building his intuition for frame-time budgets, input latency, and resource constraints before he started coding. | His ASUS ROG Strix G16 features a 240 Hz 2560×1600 display and an RTX 5060 8 GB GPU, which he uses for both gaming capture and local AI benchmarking. | The troubleshooting habits learned from configuring game mods and graphics settings transferred directly into his TypeScript, Python, and DaVinci Resolve workflows.
- **Technical Angle:** Maps 240 Hz frame-time math (4.16 ms/frame), input transfer functions, VRAM texture allocation, and state machine ticks to web UI rendering and local LLM memory management.
- **Emotional Angle:** Vindicating the curiosity of young gamers—showing how what looks like 'just playing' from the outside can actually be the birthplace of serious technical intuition.
- **Visual Sub-Blocks:** The 4-Stage Mental Shift From Player to Systems Thinker (Illustrates how curiosity transforms passive software consumption into active engineering analysis.); Casual Gaming Mindset vs. Systems Builder Mindset (Distinguishes passive entertainment consumption from analytical systems observation.); A Game Loop Is Just a State Machine With a Time Budget (Shows the structural isomorphism between a 240 Hz game loop and real-time software execution.); Four Engineering Habits I Inherited Directly From Gaming (Connects gaming discipline to measurable engineering and content creation outputs.)
- **Connected Chapters:** /blog/from-gamer-to-youtuber-building-ns-gaming-identity, /blog/games-that-shaped-my-creator-journey-free-fire-rlcraft-battlefield-cod, /blog/why-i-started-coding-from-using-software-to-building-it, /blog/my-asus-rog-strix-g16-setup-ryzen-9-rtx-5060-lab
### BLOG #04: From Gamer to YouTuber: How NS GAMING Turned Gameplay Into a Creative Studio
- **URL:** https://naborajs.dev/blog/from-gamer-to-youtuber-building-ns-gaming-identity
- **Category:** YouTube & Creator Studio | **Timeline:** 2021–2026 · Building NS GAMING (@Nishant_sarkar)
- **Summary:** Pressing record for the first time changes how you play a game—suddenly you aren't just reacting to the match, you're thinking about pacing, voice, thumbnails, and the viewer on the other side of the screen. — Treating NS GAMING not as a dump of match recordings, but as a creative production pipeline—combining 2560×1600 capture, tight narrative cuts in DaVinci Resolve, and active community hubs. — Developing a repeatable creator workflow across YouTube (@Nishant_sarkar), Telegram, WhatsApp, and Discord that survived major setbacks and sharpened my eye for media quality.
- **Documented Facts:** NS GAMING (https://youtube.com/@Nishant_sarkar) is Naboraj Sarkar's gaming and creator brand operated under the name Nishant Sarkar. | His video production pipeline captures gameplay at 2560×1600 on an ASUS ROG Strix G16 and masters exports at 4K (3840×2160) in DaVinci Resolve. | He maintains direct community channels alongside YouTube via Telegram (https://t.me/nsgamming69), WhatsApp Channel, and Discord (https://discord.gg/g7RC5zgmDy).
- **Technical Angle:** Details the media pipeline from 2560×1600 (16:10) native capture to 3840×2160 (16:9) DaVinci Resolve timelines, audio ducking, and multi-platform distribution.
- **Emotional Angle:** Captures the vulnerability of recording your own voice for the first time and the pride of turning a bedroom desk into a disciplined 4K production studio.
- **Visual Sub-Blocks:** The 4-Stage NS GAMING Production Pipeline (Documents the repeatable media engineering workflow behind NS GAMING (@Nishant_sarkar).); Raw Gameplay Upload vs. Studio-Edited NS GAMING Video (Contrasts unedited gameplay dumps with intentional creator storytelling.); My Pre-Upload Quality Checklist for @Nishant_sarkar (Shows disciplined quality control in a student creator's workflow.); Pre-Publish Metadata & Community Link Template (Provides the verified social and community routing used across NS GAMING uploads.)
- **Connected Chapters:** /blog/growing-up-with-gaming-how-play-turned-into-systems-thinking, /blog/what-losing-a-50k-subscriber-youtube-channel-taught-me, /blog/why-i-care-about-video-quality-2560x1600-to-4k-davinci-resolve, /blog/subscriber-counts-are-weird-vanity-metrics-vs-real-community
### BLOG #05: What Losing a 50K-Subscriber YouTube Channel Taught Me About Ownership and Resilience
- **URL:** https://naborajs.dev/blog/what-losing-a-50k-subscriber-youtube-channel-taught-me
- **Category:** YouTube & Creator Studio | **Timeline:** The 50K Milestone & The Rebuild Turning Point
- **Summary:** Losing a YouTube channel that had reached around 50,000 subscribers teaches you something no creator tutorial ever prepares you for: the difference between renting an audience and owning your craft. — Realizing that while a platform can remove a channel URL, it cannot erase your editing muscle memory, your voice, your technical knowledge, or the direct hubs you own. — Rebuilding NS GAMING (@Nishant_sarkar), expanding into full-stack web engineering, and architecting this self-hosted personal website so my core archive always belongs to me.
- **Documented Facts:** Naboraj Sarkar (Nishant / NS GAMING) previously built a YouTube channel to around 50,000 subscribers before losing it. | Following the loss, he rebuilt his YouTube presence at https://youtube.com/@Nishant_sarkar and expanded direct community hubs on Telegram, WhatsApp Channel, and Discord. | The experience of losing a ~50K channel directly motivated him to build his own personal website and document his work in code he controls.
- **Technical Angle:** Applies distributed systems thinking (eliminating single points of failure, canonical self-hosted hubs, multi-node broadcast redundancy) to creator media architecture.
- **Emotional Angle:** Candid without self-pity; acknowledges the real sting of losing years of visible progress while focusing on the quiet confidence of retained skill.
- **Visual Sub-Blocks:** Rented Platform Metrics vs. Owned Creator Capabilities (Distinguishes fragile platform-hosted vanity metrics from durable human capital and owned web infrastructure.); From Losing 50K to Designing a Multi-Hub Ecosystem (Traces the strategic pivot from single-channel reliance to a decentralized creator + developer network.); The Two Zeros Are Never Equal (Encapsulates the psychological breakthrough that separates experienced rebuilders from first-time beginners.); Decentralized Creator Redundancy Config (Shows how software redundancy principles apply to creator distribution.)
- **Connected Chapters:** /blog/starting-over-when-everyone-thinks-you-already-made-it, /blog/subscriber-counts-are-weird-vanity-metrics-vs-real-community, /blog/from-gamer-to-youtuber-building-ns-gaming-identity, /blog/building-my-personal-website-ui-seo-aeo-and-identity, /blog/what-my-failed-experiments-taught-me-models-repos-and-rebuilds
### BLOG #06: Starting Over When People Think You Already Made It: The Psychology of Rebuilding Online
- **URL:** https://naborajs.dev/blog/starting-over-when-everyone-thinks-you-already-made-it
- **Category:** YouTube & Creator Studio | **Timeline:** The Rebuild Era · Siliguri & Online Communities
- **Summary:** Starting from zero for the first time is exciting because nobody expects anything from you. Starting over after you've already built a 50K community is a completely different psychological test. — Detaching my ego from comparing new uploads against old peak numbers, and focusing entirely on making every new video and software project higher quality than anything I made before. — Turning the rebuild into a masterclass in quiet discipline—where craft quality, multi-platform community, and new engineering skills outgrew my old identity.
- **Documented Facts:** After losing his ~50,000-subscriber channel, Naboraj Sarkar rebuilt his YouTube presence on https://youtube.com/@Nishant_sarkar while maintaining 4K DaVinci Resolve production standards. | He used the rebuild period to broaden his public work beyond gaming alone into full-stack web development, automation (E.D.I.T.H.), and local LLM experimentation. | He stays connected with his core community across Telegram (t.me/nsgamming69), WhatsApp Channel, Discord, Instagram (@naborajs), and X (@ItsNaborajs).
- **Technical Angle:** Frames creator psychology as an input/output feedback loop—isolating controllable engineering and editorial inputs from stochastic platform impression variance.
- **Emotional Angle:** Deeply honest about the awkwardness of explaining a setback to classmates and online peers, balanced by quiet pride in refusing to cut corners on quality.
- **Visual Sub-Blocks:** First-Time Beginner vs. Experienced Rebuilder (Contrasts the emotional and technical realities of a first launch versus a post-50K rebuild.); My 4-Step Protocol for Beating the 'Nostalgia Trap' (Provides a practical cognitive framework for creators and developers recovering from a lost project or channel.); Controllable Inputs vs. Uncontrollable Outcomes (Reinforces process-oriented execution over outcome anxiety.); Evaluating Progress by Craft Delta Instead of Raw Reach (Translates rebuild psychology into deterministic evaluation logic.)
- **Connected Chapters:** /blog/what-losing-a-50k-subscriber-youtube-channel-taught-me, /blog/subscriber-counts-are-weird-vanity-metrics-vs-real-community, /blog/instagram-telegram-and-attention-social-media-as-creator-and-student, /blog/what-my-failed-experiments-taught-me-models-repos-and-rebuilds
### BLOG #07: Subscriber Counts Are Weird: What 2K, 10K, 30K, and 50K Milestones Actually Mean
- **URL:** https://naborajs.dev/blog/subscriber-counts-are-weird-vanity-metrics-vs-real-community
- **Category:** YouTube & Creator Studio | **Timeline:** 2021–2026 · Across Growth & Rebuild Milestones
- **Summary:** When you chase subscriber milestones—2K, 5K, 10K, 15K, 30K—you think each number will unlock a new version of confidence, until you realize numbers can motivate you and distract you at the exact same time. — Seeing every milestone from both sides—climbing to ~50K and then setting fresh 2K, 5K, 10K, 15K, and 30K targets during the rebuild—taught me which metrics actually reflect connection. — Using milestone targets purely as playful sprints for the community while measuring real success by watch retention, craft improvements, and active discussions in Telegram, WhatsApp, and Discord.
- **Documented Facts:** Naboraj Sarkar (Nishant / NS GAMING) has experienced both reaching ~50,000 subscribers on a legacy channel and progressing through 2K, 5K, 10K, 15K, and 30K milestone goals during his channel rebuild. | He uses public milestone checklists on https://youtube.com/@Nishant_sarkar as shared community checkpoints while evaluating true channel health through watch retention and returning viewers. | He maintains active community conversations outside algorithmic feeds via Telegram (https://t.me/nsgamming69), WhatsApp Channel, and Discord (https://discord.gg/g7RC5zgmDy).
- **Technical Angle:** Contrasts cumulative stock metrics (total subscribers) with flow and cohort metrics (first-30s retention, average percentage viewed, returning viewer ratio, and off-platform community activity).
- **Emotional Angle:** Grounded, self-aware, and freeing—helping fellow creators and students step off the hedonic treadmill of dashboard refreshing.
- **Visual Sub-Blocks:** What Each Milestone Tier Actually Unlocks in Your Craft (Reframes subscriber milestones (2K, 5K, 10K, 15K, 30K, 50K) as skill-validation checkpoints rather than status symbols.); Vanity Metrics vs. High-Signal Community Metrics (Distinguishes cumulative historical counters from live engagement and craft quality indicators.); The Four Traps of Milestone Obsession (Documents the behavioral pitfalls of metric fixation for young creators.); Community Health Score Formula (Expresses Naboraj Sarkar's philosophy of community depth over vanity metrics in code.)
- **Connected Chapters:** /blog/what-losing-a-50k-subscriber-youtube-channel-taught-me, /blog/starting-over-when-everyone-thinks-you-already-made-it, /blog/instagram-telegram-and-attention-social-media-as-creator-and-student, /blog/from-gamer-to-youtuber-building-ns-gaming-identity
### BLOG #08: Inside My ASUS ROG Strix G16 Setup: How an AMD Ryzen 9 & RTX 5060 Laptop Became My Lab
- **URL:** https://naborajs.dev/blog/my-asus-rog-strix-g16-setup-ryzen-9-rtx-5060-lab
- **Category:** Gaming & Systems Thinking | **Timeline:** 2025–2026 · Primary Hardware Workstation in Siliguri
- **Summary:** My ASUS ROG Strix G16—running an AMD Ryzen 9 8940HX, an NVIDIA RTX 5060 Laptop GPU with 8 GB VRAM, 16 GB DDR5 RAM, and a 240 Hz 16-inch display—isn't just for gaming; it's where I hit my first real hardware limits in AI and 4K rendering. — Treating those hardware boundaries as an engineering classroom—learning how CPU core scheduling, NVENC media engines, VRAM quantization, and DDR5 memory offloading actually work. — Turning a single laptop on my desk in Siliguri into a three-in-one workstation that handles 2560×1600 gaming capture, 4K video production, and local AI development.
- **Documented Facts:** Naboraj Sarkar's primary workstation is an ASUS ROG Strix G16 laptop. | Its exact specifications are an AMD Ryzen 9 8940HX processor, NVIDIA GeForce RTX 5060 Laptop GPU with 8 GB VRAM, 16 GB DDR5 system RAM, and a 16-inch 2560×1600 (16:10) 240 Hz display. | He uses this machine for three main workloads: 2560×1600 gameplay capture, 4K (3840×2160) video editing in DaVinci Resolve, and local AI / LLM benchmarking.
- **Technical Angle:** Covers mobile workstation architecture: AMD Ryzen 9 8940HX multi-core compute, RTX 5060 8 GB GDDR memory ceilings, 16 GB DDR5 PCIe offload bottlenecks, and 2560×1600 (16:10) 240 Hz display math.
- **Emotional Angle:** Appreciation for a hard-working machine that serves as bedroom gaming rig, film studio, and AI research lab all at once.
- **Visual Sub-Blocks:** Core Hardware Specs of My ASUS ROG Strix G16 Lab (Provides the canonical hardware specifications of Naboraj Sarkar's primary workstation.); Where the ROG Strix G16 Flies vs. Where It Hits Physical Ceilings (Evaluates real-world performance envelopes across gaming, 4K video editing, and local LLM inference.); VRAM Budget Calculator for My 8 GB RTX 5060 Lab (Demonstrates practical memory budgeting on an 8 GB VRAM laptop GPU.); How One Laptop Switches Across Three Daily Modes (Shows operational discipline in managing thermals, memory, and display modes on a single machine.)
- **Connected Chapters:** /blog/why-i-care-about-video-quality-2560x1600-to-4k-davinci-resolve, /blog/24-billion-parameters-versus-my-8gb-vram-hardware-wall, /blog/running-ai-models-on-my-own-hardware-the-local-llm-rabbit-hole, /blog/games-that-shaped-my-creator-journey-free-fire-rlcraft-battlefield-cod
### BLOG #09: Why I Obsess Over Video Quality: Capturing at 2560×1600 and Mastering 4K in DaVinci Resolve
- **URL:** https://naborajs.dev/blog/why-i-care-about-video-quality-2560x1600-to-4k-davinci-resolve
- **Category:** YouTube & Creator Studio | **Timeline:** 2023–2026 · Media Engineering & DaVinci Resolve Pipeline
- **Summary:** Anyone can upload raw compressed gameplay, but learning how 2560×1600 capture, frame pacing, bitrate math, and 3840×2160 4K timelines in DaVinci Resolve interact changed how I see visual engineering. — Designing a clean DaVinci Resolve mastering pipeline on my ASUS ROG Strix G16 that maps 2560×1600 source files into crisp 3840×2160 4K exports with consistent frame pacing and high-bitrate encoding. — Every upload on @Nishant_sarkar preserves fine foliage, crosshair detail, and fast camera motion—while teaching me visual precision that directly inspired my web UI design.
- **Documented Facts:** Naboraj Sarkar captures gameplay at native 2560×1600 resolution on his ASUS ROG Strix G16's 16-inch 240 Hz display. | He edits and masters his videos in DaVinci Resolve on a 3840×2160 (4K UHD) timeline to preserve motion clarity and unlock higher streaming bitrates on YouTube (@Nishant_sarkar). | He optimizes his DaVinci Resolve node trees and render cache to stay within the 8 GB VRAM ceiling of his RTX 5060 Laptop GPU.
- **Technical Angle:** Explains 16:10 (`2560×1600`) to 16:9 (`3840×2160`) 1.5× horizontal scaling math, 240 Hz to 60 fps Constant Frame Rate (4:1 cadence) synchronization, inter-frame compression macroblocking, and 8 GB VRAM memory management in DaVinci Resolve.
- **Emotional Angle:** Craftsman's pride—finding genuine joy in tuning frame cadences, color curves, and export bitrates until fast gameplay looks like cinema.
- **Visual Sub-Blocks:** The 4-Stage 2560×1600 to 4K DaVinci Resolve Pipeline (Documents the end-to-end resolution, aspect-ratio, and bitrate pipeline used by NS GAMING.); Default 1080p Upload vs. 2560×1600 → 4K DaVinci Resolve Master (Explains the technical advantage of 4K mastering for high-motion gaming content.); Aspect Ratio & Scaling Math: 2560×1600 (16:10) to 3840×2160 (16:9) (Shows the exact mathematical relationship between 2560×1600 capture and 3840×2160 4K output.); Keeping 4K DaVinci Resolve Smooth on an 8 GB VRAM GPU (Connects media production optimization back to the 8 GB VRAM constraint of the ASUS ROG Strix G16.)
- **Connected Chapters:** /blog/my-asus-rog-strix-g16-setup-ryzen-9-rtx-5060-lab, /blog/from-gamer-to-youtuber-building-ns-gaming-identity, /blog/games-that-shaped-my-creator-journey-free-fire-rlcraft-battlefield-cod, /blog/my-dream-of-making-a-cinematic-website-neon-ui-and-motion
### BLOG #10: The Games That Shaped My Creator Journey: Free Fire, Minecraft RLCraft, Roblox, Battlefield & Call of Duty
- **URL:** https://naborajs.dev/blog/games-that-shaped-my-creator-journey-free-fire-rlcraft-battlefield-cod
- **Category:** Gaming & Systems Thinking | **Timeline:** 2020–2026 · Five Titles That Defined NS GAMING
- **Summary:** Every game I've recorded or played seriously—from Free Fire sensitivity tuning to brutal RLCraft survival, Roblox FPS mechanics, Battlefield chaos, and Call of Duty campaigns—taught me a different lesson about pacing and audience psychology. — Decoding the unique mechanical and narrative DNA of each title I played on NS GAMING, and adapting my capture settings, commentary cadence, and DaVinci Resolve cuts to match. — Building a versatile creator toolkit that understands precision mechanics, underdog survival arcs, fast arena loops, large-scale spectacle, and cinematic storytelling.
- **Documented Facts:** The core games that shaped Naboraj Sarkar's (Nishant / NS GAMING) creator journey include Free Fire, Minecraft (specifically the RLCraft modpack), Roblox FPS/competitive modes, Battlefield, and Call of Duty. | Playing and editing Free Fire taught him input sensitivity physics, DPI tuning, and fast montage pacing. | Minecraft RLCraft taught him modpack configuration, JVM resource management, and survival storytelling, while Battlefield and Call of Duty served as 2560×1600 to 4K DaVinci Resolve visual benchmarks on his ASUS ROG Strix G16.
- **Technical Angle:** Covers X/Y input sensitivity and eDPI transfer functions (Free Fire / FPS), Java Virtual Machine memory and mod dependency graphs (Minecraft RLCraft), 240 Hz arena loop latency (Roblox), and high-entropy 4K video encoding (Battlefield & Call of Duty).
- **Emotional Angle:** Nostalgic yet analytical—celebrating the specific matches, modpacks, and campaigns that turned a gaming hobby in Siliguri into a disciplined creative craft.
- **Visual Sub-Blocks:** Four Mechanical & Creative Lessons From My Core Game Roster (Maps specific game titles (Free Fire, RLCraft, Roblox, Battlefield/CoD) to concrete engineering and media lessons.); Editing Reflex Shooters vs. Editing Survival & Cinematic Games (Demonstrates genre-aware video editing and audience psychology.); Sensitivity & Effective Edpi / Angular Transfer Model (Connects legacy Free Fire sensitivity physics and FPS mechanics to mathematical modeling.); How Gaming Completed Part 1 and Opened the Door to Coding (Synthesizes Blogs 01–10 and sets up Part 2 (Coding, Automation & Local AI).)
- **Connected Chapters:** /blog/growing-up-with-gaming-how-play-turned-into-systems-thinking, /blog/from-gamer-to-youtuber-building-ns-gaming-identity, /blog/why-i-care-about-video-quality-2560x1600-to-4k-davinci-resolve, /blog/my-asus-rog-strix-g16-setup-ryzen-9-rtx-5060-lab
### BLOG #11: Why I Started Coding: The Shift From Using Technology to Building My Own
- **URL:** https://naborajs.dev/blog/why-i-started-coding-from-using-software-to-building-it
- **Category:** Coding & Systems | **Timeline:** 2022–2024 • Consumer-to-Builder Transition
- **Summary:** At some point, just playing games and using apps stopped being enough—I wanted to open the hood and understand how data, servers, and interfaces actually worked. — Writing my first HTML/JS files and inspecting network payloads made me realize every digital product is just human-written logic. — Today, under Naboraj Sarkar (github.com/naborajs) and Nishant Sarkar, my default reflex when facing a problem is to architect my own tool.
- **Documented Facts:** Naboraj Sarkar (born 19 August 2010 in Siliguri, West Bengal, India) is a Class 10 CBSE student who builds software under his real name and creator alias Nishant Sarkar. | His public code and web platform work are published at https://github.com/naborajs. | His early creative background began with gaming and media production on YouTube (@Nishant_sarkar) before expanding into full-stack web and AI engineering.
- **Technical Angle:** Explains the fundamental client-server-state loop (DOM events, local state, HTTP payloads) that demystifies modern applications.
- **Emotional Angle:** Restless curiosity transforming into quiet confidence as intimidating interfaces become understandable code.
- **Visual Sub-Blocks:** The F12 Inspector Moment (Represents the psychological transition from passive digital consumption to active technical curiosity.); Using Apps vs. Engineering Systems (Contrasts surface-level app usage with structural systems thinking.); My 4-Stage Progression Into Software Engineering (Maps the sequential skill acquisition of Naboraj Sarkar from gamer to full-stack builder.); The Core Loop of Every App I Build (Illustrates Naboraj's focus on typed, predictable software loops.)
- **Connected Chapters:** /blog/the-first-time-coding-stopped-feeling-like-homework, /blog/my-obsession-with-building-websites-as-technical-playgrounds, /blog/growing-up-with-gaming-how-play-turned-into-systems-thinking, /blog/who-is-naboraj-sarkar-nishant-student-gamer-builder
### BLOG #12: The First Time Coding Stopped Feeling Like Homework and Started Feeling Like Leverage
- **URL:** https://naborajs.dev/blog/the-first-time-coding-stopped-feeling-like-homework
- **Category:** Coding & Systems | **Timeline:** 2023–2024 • The Leverage Awakening
- **Summary:** Syntax drills in a textbook feel like memorization, but the moment a script you wrote solves a real problem on your own screen, coding turns into a completely different addiction. — Building interactive tools I actually used for my own workflows made every loop, function, and API call feel like a superpower. — Coding became my highest-leverage habit: write logic once, and let the computer execute it thousands of times accurately.
- **Documented Facts:** Naboraj Sarkar transitioned from basic syntax exercises to building over 30 interactive browser utilities integrated into his personal web platform (github.com/naborajs/web). | He balances Class 10 CBSE academic coursework in Siliguri with self-directed software engineering. | His utilities focus on practical client-side execution, clean TypeScript state management, and zero-clutter interfaces.
- **Technical Angle:** Highlights reusable functions, deterministic state transformations, and modular utility design in TypeScript.
- **Emotional Angle:** The spark of agency and excitement when a script you wrote overnight becomes a tool you rely on the next day.
- **Visual Sub-Blocks:** The First Self-Built Utility That Stayed Open in My Browser (Shows how personal dogfooding (using your own software) accelerates engineering maturity.); Textbook Exercises vs. Real-World Engineering (Distinguishes rote syntax memorization from applied software engineering.); The 4-Step Leverage Loop I Still Use Today (Codifies Naboraj's repeatable product-building loop.); Turning Repetitive Work Into a Reusable Function (Concrete example of code leverage inside Naboraj's blog engine.)
- **Connected Chapters:** /blog/why-i-started-coding-from-using-software-to-building-it, /blog/my-obsession-with-building-websites-as-technical-playgrounds, /blog/what-school-teaches-that-self-taught-coding-doesnt, /blog/the-day-i-started-taking-ai-seriously-beyond-chatbots
### BLOG #13: My Obsession With Building Websites: Why the Browser Is My Favorite Laboratory
- **URL:** https://naborajs.dev/blog/my-obsession-with-building-websites-as-technical-playgrounds
- **Category:** Web Architecture | **Timeline:** 2024–2026 • The Web Playground Era
- **Summary:** A website is the fastest way to turn an idea in your head into something anyone in the world can click, test, and experience within seconds. — Realizing the modern browser is a full operating system capable of 60fps animations, local storage, audio synthesis, Canvas rendering, and live AI streaming. — Every experiment I build—from interactive knowledge graphs to AI control decks—lives inside an instant-access web playground.
- **Documented Facts:** Naboraj Sarkar uses his personal repository (github.com/naborajs/web) as an active technical sandbox combining a 50-blog knowledge base, 30+ browser utilities, and AI integrations. | The platform is built with React, TypeScript, Tailwind CSS, Express, and Supabase. | He prioritizes zero-install browser tools so users on any device can test his software immediately.
- **Technical Angle:** Examines client-side execution, Vite hot-module replacement, shared TypeScript schemas, and browser runtime capabilities.
- **Emotional Angle:** Pure creative joy from having a medium where visual design, interactive motion, and deep systems code merge instantaneously.
- **Visual Sub-Blocks:** Hot-Reloading at Midnight in Siliguri (Captures why tight feedback loops accelerate learning and UI craftsmanship.); 4 Layers Inside Every Web Playground I Build (Outlines the modular stack Naboraj uses for web experimentation.); Static Brochure Site vs. Technical Playground Site (Explains why Naboraj continuously expands his personal platform rather than leaving it static.); Zero-Latency Client-Side Sandbox Pattern (Shows how client-side execution delivers native-like speed in the browser.)
- **Connected Chapters:** /blog/building-my-personal-website-ui-seo-aeo-and-identity, /blog/when-a-website-becomes-more-than-a-website-platform-architecture, /blog/my-dream-of-making-a-cinematic-website-neon-ui-and-motion, /blog/why-i-started-coding-from-using-software-to-building-it
### BLOG #14: Architecting My Personal Website: Balancing Dark/Light Themes, SEO, AEO, and Identity
- **URL:** https://naborajs.dev/blog/building-my-personal-website-ui-seo-aeo-and-identity
- **Category:** Web Architecture | **Timeline:** 2025–2026 • Flagship Web Architecture
- **Summary:** When I started building github.com/naborajs/web, I didn't just want a static portfolio—I wanted a living system engineered for humans, Google Search, and AI answer engines alike. — Designing a unified architecture that pairs adaptive blue-accented Dark/Light themes for humans with structured JSON-LD, documented facts, and bidirectional wiki-links for search and AI engines. — My website now serves three audiences simultaneously: human visitors, traditional search crawlers (SEO), and AI answer engines (AEO).
- **Documented Facts:** The personal website of Naboraj Sarkar (Nishant Sarkar) is hosted in the public repository https://github.com/naborajs/web. | The platform implements both SEO (Search Engine Optimization) and AEO (Answer Engine Optimization) through structured `documentedFacts`, `quickFacts`, and bidirectional `connectedSlugs`. | The interface features an adaptive dark/light theme system designed around Naboraj's favorite color, blue. | Entity metadata explicitly connects Naboraj Sarkar with his online aliases Nishant Sarkar and NS GAMING (@Nishant_sarkar).
- **Technical Angle:** Covers CSS variable theming, entity disambiguation, JSON-LD/AEO structured facts, and bidirectional knowledge graph schemas in TypeScript.
- **Emotional Angle:** Pride in craftsmanship—building a digital home that looks refined to human eyes while remaining crystal-clear to search and AI systems.
- **Visual Sub-Blocks:** The Triple-Audience Web Architecture (Explains the architectural philosophy behind Naboraj's personal web platform.); Traditional SEO vs. Answer Engine Optimization (AEO) (Demonstrates Naboraj's forward-looking approach to web discoverability.); How My Dual Dark/Light Theme Engine Works (Highlights Naboraj's attention to visual contrast, accessibility, and personal aesthetic (favorite color: blue).); Entity & AEO Metadata Structure in TypeScript (Shows the exact TypeScript pattern powering the 50-blog knowledge engine.)
- **Connected Chapters:** /blog/when-a-website-becomes-more-than-a-website-platform-architecture, /blog/my-dream-of-making-a-cinematic-website-neon-ui-and-motion, /blog/why-i-keep-renaming-my-projects-ns-gaming-devspace-codex-nsgo, /blog/who-is-naboraj-sarkar-nishant-student-gamer-builder
### BLOG #15: Why I Kept Renaming My Projects: From NS GAMING to NS DevSpace, NS Codex, NSGO, and Naborajs
- **URL:** https://naborajs.dev/blog/why-i-keep-renaming-my-projects-ns-gaming-devspace-codex-nsgo
- **Category:** Nishant's Journey | **Timeline:** 2020–2026 • Brand & System Evolution
- **Summary:** If you look through my project history—NS GAMING, NS DevSpace, NS Codex, NSGO, EDITH, Naborajs—it looks like I couldn't make up my mind, when in reality each name was a snapshot of what I was learning to build next. — Realizing that renaming wasn't indecision—it was version control for my own identity as I moved from gaming content to developer workspaces, local AI endpoints, and unified platforms. — Today, every past name sits in a clear lineage under Naboraj Sarkar (naborajs) and Nishant Sarkar.
- **Documented Facts:** Naboraj Sarkar's project naming lineage spans NS GAMING, NS DevSpace, NS Codex, NSGO, EDITH, and Naborajs. | The 'NS' prefix stands for both Naboraj Sarkar (legal/developer name) and Nishant Sarkar (online creator identity). | NS DevSpace and NS Codex represent his web tool and knowledge-base phases, while NSGO and EDITH represent his local/cloud AI endpoint and WhatsApp business AI agent systems. | His unified public code portfolio is hosted at https://github.com/naborajs.
- **Technical Angle:** Explains modular branding, config-driven site metadata, legacy slug resolution, and subsystem separation.
- **Emotional Angle:** Affectionate honesty about the ambitious codenames a young developer invents while leveling up from bedroom gamer to systems builder.
- **Visual Sub-Blocks:** The 5-Era Evolution of the 'NS' Ecosystem (Provides a canonical entity map for search engines and readers encountering any of Naboraj's project names.); Why Codenames Made Me Build Harder (Explains the psychological value of product branding during self-taught engineering.); Hardcoded Branding vs. Config-Driven Identity (Shows how personal rebranding led to better DRY (Don't Repeat Yourself) engineering practices.); Canonical Brand & Project Lineage Map (Provides a clear, machine-readable reference of the NS naming lineage.)
- **Connected Chapters:** /blog/why-i-became-nishant-online-identity-vs-creator-branding, /blog/building-my-personal-website-ui-seo-aeo-and-identity, /blog/what-cloudflare-tunnels-taught-me-about-networking-and-apis, /blog/what-my-failed-experiments-taught-me-models-repos-and-rebuilds
### BLOG #16: When a Website Becomes More Than a Website: Connecting React, Vercel, Supabase, and AI APIs
- **URL:** https://naborajs.dev/blog/when-a-website-becomes-more-than-a-website-platform-architecture
- **Category:** Web Architecture | **Timeline:** 2025–2026 • Full-Stack Platform Engineering
- **Summary:** The moment you add authentication, a Supabase database, an admin control deck, 30+ interactive browser tools, and an AI assistant, your personal site stops being a brochure and becomes a software platform. — Architecting a strict 3-folder monorepo style (`client/`, `server/`, `shared/`) with unified TypeScript schemas, edge-ready API routes, and resilient database fallbacks. — My personal website now operates like a mini SaaS platform—fast, type-safe, and extensible.
- **Documented Facts:** Naboraj Sarkar's web platform (github.com/naborajs/web) integrates a React + TypeScript frontend, Express/Vercel API routes, and a Supabase PostgreSQL database. | The platform includes over 30 interactive browser tools, a 50-post bidirectional knowledge wiki, an authenticated admin control deck, and AI assistant integrations. | A shared TypeScript layer (`shared/`) enforces type safety across both client and server modules.
- **Technical Angle:** Covers monorepo-style `client/shared/server` separation, hybrid static-plus-Supabase data loading, serverless API routing on Vercel, and type-safe schemas.
- **Emotional Angle:** The shift from feeling like a hobbyist coder to feeling like a systems architect orchestrating frontend, backend, database, and AI layers.
- **Visual Sub-Blocks:** End-to-End Request Lifecycle on My Platform (Illustrates the full-stack engineering behind github.com/naborajs/web.); Why I Chose a Hybrid Static + Database Architecture (Explains the hybrid data strategy used in Naboraj's platform.); The 4 Core Subsystems Running Inside One Repo (Breaks down the functional modules of the platform.); Resilient Hybrid Data Loader Pattern (Demonstrates defensive full-stack engineering.)
- **Connected Chapters:** /blog/building-my-personal-website-ui-seo-aeo-and-identity, /blog/why-i-love-building-dashboards-turning-chaos-into-control, /blog/the-idea-behind-edith-whatsapp-ai-agent-by-ns, /blog/my-obsession-with-building-websites-as-technical-playgrounds
### BLOG #17: My Dream of Making a Cinematic Website: Neon Typography, 3D Motion, and Title-Sequence Intros
- **URL:** https://naborajs.dev/blog/my-dream-of-making-a-cinematic-website-neon-ui-and-motion
- **Category:** UI & Motion Design | **Timeline:** 2024–2026 • Cinematic UI & Motion Craft
- **Summary:** Most developer websites look like plain resumes, but growing up with gaming and video editing made me want a website that opens with the energy of a cinematic title sequence. — Applying video-editing pacing from DaVinci Resolve to CSS/Framer Motion—using GPU-accelerated transforms, electric blue glow tokens, and optional/skippable intro choreography. — An interface that feels like stepping into a sci-fi control deck while remaining fast, readable, and responsive.
- **Documented Facts:** Naboraj Sarkar's favorite color is blue, which serves as the primary visual accent across his web platform (github.com/naborajs/web). | His UI design combines influences from gaming HUDs and DaVinci Resolve video editing with React and Tailwind CSS engineering. | The blog engine in `shared/blog-types.ts` supports configurable `BlogAnimationPreset` and `BlogVisualTheme` schemas for studio-style visual explainers.
- **Technical Angle:** Explains GPU-composited animations (`transform`, `opacity`), spring physics parameters, CSS radial glow tokens, and reduced-motion ergonomics.
- **Emotional Angle:** Creative ambition—refusing to settle for a plain template and instead chasing the goosebumps of a great game or movie intro on the web.
- **Visual Sub-Blocks:** Thinking About Web Pages Like a DaVinci Resolve Timeline (Shows how Naboraj's video-editing background directly informs his front-end UI engineering.); Boring Resume Template vs. Cinematic Engineering Portal (Contrasts cookie-cutter portfolios with Naboraj's expressive design system.); My 4 Rules for 60fps Cinematic Web Motion (Documents the engineering constraints Naboraj applies to UI animations.); Configurable Animation Presets in My Blog Engine (Shows how cinematic motion is systematized in TypeScript rather than hacked together.)
- **Connected Chapters:** /blog/building-my-personal-website-ui-seo-aeo-and-identity, /blog/why-i-care-about-video-quality-2560x1600-to-4k-davinci-resolve, /blog/my-obsession-with-building-websites-as-technical-playgrounds, /blog/school-projects-but-make-them-look-like-a-product-launch
### BLOG #18: Why I Love Building Dashboards: Turning Messy Systems Into Organized Interfaces
- **URL:** https://naborajs.dev/blog/why-i-love-building-dashboards-turning-chaos-into-control
- **Category:** Systems & UI | **Timeline:** 2025–2026 • Control Deck Architecture
- **Summary:** There is a specific kind of satisfaction in taking a chaotic pile of logs, database rows, and API states and organizing them into a clean dashboard where everything makes visual sense. — Designing structured control decks with status telemetry, search/filter pipelines, and instant action triggers. — From my website's Admin Studio to EDITH's 10-module AI command center, dashboards became my signature way of making complex backends human-controllable.
- **Documented Facts:** Naboraj Sarkar engineered custom control dashboards for both his personal platform (`github.com/naborajs/web`) and his WhatsApp AI agent system (EDITH). | His dashboard architecture emphasizes semantic status colors, KPI telemetry cards, instant client-side filtering, and type-safe mutations. | Every blog post in his 50-post series includes a structured `BlogAnalyticsChart` component displaying quantitative metrics.
- **Technical Angle:** Covers telemetry aggregation, semantic color tokens, client-side search/filter pipelines, and validated CRUD interfaces in React and TypeScript.
- **Emotional Angle:** The calm sense of mastery that comes from taming complex, noisy software states into an orderly visual workspace.
- **Visual Sub-Blocks:** The Moment a Messy Backend Gets a Clean Control Deck (Explains the psychological and practical drive behind Naboraj's dashboard architecture.); Anatomy of a High-Signal Dashboard Interface (Documents Naboraj's repeatable layout pattern for control interfaces.); Cluttered Data Dump vs. Engineered Control Interface (Distinguishes amateur admin templates from purposeful operational interfaces.); Typed Telemetry & KPI Aggregator (Shows the clean TypeScript aggregation logic behind dashboard summary cards.)
- **Connected Chapters:** /blog/the-dashboard-behind-edith-10-modules-for-business-ai, /blog/when-a-website-becomes-more-than-a-website-platform-architecture, /blog/building-my-personal-website-ui-seo-aeo-and-identity, /blog/the-idea-behind-edith-whatsapp-ai-agent-by-ns
### BLOG #19: The Day I Started Taking AI Seriously: Moving Beyond Chat Windows to System Architecture
- **URL:** https://naborajs.dev/blog/the-day-i-started-taking-ai-seriously-beyond-chatbots
- **Category:** AI & Systems | **Timeline:** 2024–2026 • The AI Engineering Shift
- **Summary:** At first, AI was just a browser tab where you typed questions. Everything changed for me the day I started asking how the model weights, context windows, and APIs underneath actually worked. — Learning to treat LLMs as programmable reasoning engines wrapped inside strict TypeScript/Zod schemas, system prompts, and tool-calling pipelines. — AI stopped being a novelty tab in my browser and became a core architectural layer powering local Ollama experiments, NSGO endpoints, and EDITH.
- **Documented Facts:** Naboraj Sarkar (Nishant Sarkar) transitioned from using web-based AI chat interfaces to engineering programmatic AI systems using TypeScript, OpenAI-compatible `/v1` APIs, and local Ollama models. | His TypeSafe AI pattern validates LLM JSON responses at runtime with deterministic fallbacks to prevent application crashes. | This architectural shift laid the foundation for his NSGO AI gateway and EDITH WhatsApp business AI agent platform.
- **Technical Angle:** Covers token context construction, temperature control, JSON sanitization, TypeScript/Zod runtime validation, and deterministic fallback handlers.
- **Emotional Angle:** The intellectual thrill of peeling back the hype around AI and discovering clean, controllable engineering principles underneath.
- **Visual Sub-Blocks:** TypeSafe AI: Kya Hai & How It Works Under the Hood (Explains the core TypeSafe AI architecture pattern Naboraj uses across his AI projects (and preserves the legacy concept from `typesafe-ai-architecture-kya-hai-live-test`).); Unstructured Prompting vs. Engineered AI Systems (Contrasts naive LLM integration with production-grade AI system design.); My 4-Layer Mental Model of Any Modern AI Application (Shows Naboraj's structural decomposition of AI software.); TypeSafe AI Response Parser With Fallback (Demonstrates practical TypeSafe AI engineering in TypeScript.)
- **Connected Chapters:** /blog/running-ai-models-on-my-own-hardware-the-local-llm-rabbit-hole, /blog/i-didnt-want-a-chatbot-i-wanted-an-ai-workflow-agent, /blog/the-idea-behind-edith-whatsapp-ai-agent-by-ns, /blog/why-i-started-coding-from-using-software-to-building-it
### BLOG #20: Running AI Models on My Own Hardware: Why Ollama and Local LLMs Hooked Me
- **URL:** https://naborajs.dev/blog/running-ai-models-on-my-own-hardware-the-local-llm-rabbit-hole
- **Category:** Local AI & Hardware | **Timeline:** 2025–2026 • The Local LLM Rabbit Hole
- **Summary:** The first time an AI model generates code on your own laptop with the Wi-Fi turned off, it stops feeling like magic in someone else's cloud and starts feeling like engineering on your own machine. — Installing Ollama on my ASUS ROG Strix G16 (Ryzen 9 8940HX + RTX 5060 8GB GDDR7), pulling open-weight models, and watching my own GPU tensor cores stream tokens offline. — Local inference turned my bedroom desk in Siliguri into a real AI hardware lab, teaching me quantization, VRAM budgeting, and local `/v1` API routing.
- **Documented Facts:** Naboraj Sarkar runs local Large Language Models via Ollama on his ASUS ROG Strix G16 laptop in Siliguri, West Bengal. | His hardware configuration features an AMD Ryzen 9 8940HX processor, 16GB DDR5 RAM, 1TB NVMe SSD, and an NVIDIA GeForce RTX 5060 Laptop GPU with 8GB GDDR7 VRAM. | He uses Ollama's OpenAI-compatible `/v1/chat/completions` API to test local coding and agent workflows offline at zero token cost.
- **Technical Angle:** Covers Ollama local daemon architecture, GGUF quantization, 8GB GDDR7 VRAM allocation on the RTX 5060, and OpenAI-compatible `/v1` API integration.
- **Emotional Angle:** Pure excitement and independence from turning a personal gaming/coding laptop in Siliguri into a self-contained offline AI laboratory.
- **Visual Sub-Blocks:** Turning Off Wi-Fi and Watching the GPU Spike (Captures the exact moment AI transitioned from a cloud service to local systems engineering for Naboraj.); Black-Box Cloud API vs. Transparent Local Inference (Highlights the educational and architectural value of local inference.); How a Local Model Runs on My Laptop From Command to Token (Explains the local inference lifecycle clearly for readers.); Calling Local Ollama Using Standard `/v1` TypeScript Fetch (Shows how local LLMs integrate seamlessly into Naboraj's TypeScript stack.)
- **Connected Chapters:** /blog/24-billion-parameters-versus-my-8gb-vram-hardware-wall, /blog/quantization-explained-q4-q5-q6-q8-in-real-vram-tests, /blog/building-my-own-ai-endpoint-ollama-openai-v1-and-cloud-gpus, /blog/my-asus-rog-strix-g16-setup-ryzen-9-rtx-5060-lab, /blog/the-day-i-started-taking-ai-seriously-beyond-chatbots, /blog/i-didnt-want-a-chatbot-i-wanted-an-ai-workflow-agent
### BLOG #21: 24 Billion Parameters Versus My 8 GB VRAM: What Hitting the Hardware Wall Taught Me
- **URL:** https://naborajs.dev/blog/24-billion-parameters-versus-my-8gb-vram-hardware-wall
- **Category:** AI & Engineering | **Timeline:** 2025–2026 (Local AI Hardware Experiments)
- **Summary:** Pulling a 24B-parameter model around 15 GB in size onto a laptop with 8 GB of RTX 5060 VRAM and 16 GB of system RAM is the fastest way to learn how GPU memory bandwidth actually works. — Mapping exact layer-offloading splits, KV cache allocation, and memory bus bandwidth differences between GDDR7/GDDR6 VRAM and system DDR5 RAM. — Developing a disciplined mental model for hardware-aware AI engineering—knowing when to quantize, when to shrink context, and when to offload to cloud GPUs.
- **Documented Facts:** Tested ~15 GB 24-billion-parameter weights (including Devstral-Small-2 24B) locally on an 8 GB VRAM RTX 5060 laptop with 16 GB system RAM. | Observed partial GPU layer offloading where roughly half the transformer layers reside in VRAM while the remainder spill into system DDR5 RAM. | Measured generation speed drop to ~5–8 tokens per second during split-memory inference compared to 45+ tokens/sec on full-VRAM 8B models. | Documented Windows OS + background desktop overhead leaving roughly 11.5–12.5 GB of usable system RAM before swap thrashing begins.
- **Technical Angle:** Transformer weight memory footprint calculation, GGUF partial layer offloading in llama.cpp/Ollama, VRAM vs. DDR5 memory bandwidth bottlenecks, and OS paging thresholds.
- **Emotional Angle:** The mixture of stubborn curiosity as the progress bar hits 100%, the suspense of watching Task Manager memory graphs flatline at 96%, and the clarity that comes from understanding why the hardware behaves the way it does.
- **Visual Sub-Blocks:** The Night I Pulled a 15 GB Model onto an 8 GB GPU (Demonstrates the exact moment theoretical parameter counts collide with physical laptop memory limits.); Where Every Gigabyte Actually Goes During a 24B Load (Breaks down the exact memory math of running a 15 GB model on an 8 GB VRAM + 16 GB RAM laptop.); Full-VRAM Inference vs. Split VRAM + System RAM Inference (Explains the physics of memory-bandwidth-bound autoregressive token generation.); Inspecting the GPU/CPU Offload Split in Ollama (Shows the exact diagnostic command used to verify CPU/GPU percentage splits in Ollama.)
- **Connected Chapters:** /blog/running-ai-models-on-my-own-hardware-the-local-llm-rabbit-hole, /blog/what-5-to-8-tokens-per-second-feels-like-in-local-ai, /blog/quantization-explained-q4-q5-q6-q8-in-real-vram-tests, /blog/why-bigger-ai-models-arent-always-better-speed-vs-size, /blog/my-asus-rog-strix-g16-setup-ryzen-9-rtx-5060-lab, /blog/why-100k-context-windows-became-my-new-ai-obsession, /blog/building-my-own-ai-endpoint-ollama-openai-v1-and-cloud-gpus
### BLOG #22: Why Bigger AI Models Aren't Always Better: Model Size vs. Speed, Context, and Usability
- **URL:** https://naborajs.dev/blog/why-bigger-ai-models-arent-always-better-speed-vs-size
- **Category:** AI & Engineering | **Timeline:** 2025–2026 (Model Evaluation & Workflow Design)
- **Summary:** On paper, a 24B or 31B model looks strictly superior to an 8B model—until you actually try to use it inside a live coding loop and realize latency changes how you think. — Evaluating models not by static benchmark percentages, but by 'cognitive throughput'—how fast a human + AI pair can test, fail, and refine working code. — Adopting a right-sized model strategy: nimble 7B–9B models for high-frequency local iteration, and 24B+ models only for deep architectural passes.
- **Documented Facts:** Compared 8B class models (such as Dolphin 8B and Qwen 7B/8B variants) against 24B–31B class models (such as Devstral-Small-2 24B and Gemma large variants) on the same hardware. | Observed that smaller models leave 2.5–3.0 GB of free VRAM for larger KV caches and multi-file context. | Found that two fast iterations with an 8B model + compiler verification frequently beat a single slow generation from a 24B model.
- **Technical Angle:** Time-to-first-token (TTFT), inter-token latency, KV cache VRAM reservation vs. parameter footprint, and multi-turn self-correction loops.
- **Emotional Angle:** Moving past the ego of running the biggest possible model on my laptop to the maturity of choosing the tool that actually helps me ship code faster.
- **Visual Sub-Blocks:** The Illusion of the Leaderboard Number (Shows how real-world developer experience contradicts raw benchmark rankings.); The Three-Way Tug-of-War Inside Fixed VRAM (Illustrates the three-variable constraint equation of local LLM deployment.); Two Fast Loops vs. One Slow Monologue (Demonstrates why iterative human-in-the-loop coding favors low latency over marginal one-shot accuracy.); My Personal Decision Matrix for Picking Model Sizes (Provides a practical routing rubric for local vs. hybrid model selection.)
- **Connected Chapters:** /blog/24-billion-parameters-versus-my-8gb-vram-hardware-wall, /blog/what-5-to-8-tokens-per-second-feels-like-in-local-ai, /blog/my-search-for-the-right-coding-ai-devstral-qwen-gemma-dolphin, /blog/quantization-explained-q4-q5-q6-q8-in-real-vram-tests, /blog/building-my-own-ai-endpoint-ollama-openai-v1-and-cloud-gpus
### BLOG #23: Quantization: The Strange World of Q4_K_M, Q5_K_M, Q6_K, and Q8 Through My Own Experiments
- **URL:** https://naborajs.dev/blog/quantization-explained-q4-q5-q6-q8-in-real-vram-tests
- **Category:** AI & Engineering | **Timeline:** 2025–2026 (Local LLM Quantization Benchmarks)
- **Summary:** Before I started running local models in Ollama, strings like Q4_K_M, Q5_K_M, Q6_K, and Q8 looked like random serial numbers. Now I see them as the exact trade-off curve between memory and intelligence. — Testing Q4_K_M, Q5_K_M, Q6_K, and Q8 side-by-side on the same coding prompts to see where perplexity degradation actually shows up in TypeScript and Python generation. — Mastering GGUF k-quant selection so every model fits my exact VRAM + context window budget without sacrificing reasoning quality.
- **Documented Facts:** Tested GGUF quantization tiers (`Q4_K_M`, `Q5_K_M`, `Q6_K`, and `Q8_0`) across 8B and 24B models in Ollama. | Verified that an 8B model at `Q8_0` (~8.5 GB) crosses the 8 GB RTX 5060 VRAM ceiling and spills slightly into system RAM, whereas `Q5_K_M` (~5.7 GB) and `Q6_K` (~6.6 GB) stay 100% in VRAM. | Confirmed that for a 24B model, `Q4_K_M` (~14.5–15 GB) is the highest practical quantization that can even load within a combined 8 GB VRAM + 16 GB RAM envelope.
- **Technical Angle:** FP16 vs. integer quantization, GGUF K-quants (`_K_S`, `_K_M`, `_K_L`), mixed-precision super-blocks for attention/feed-forward weights, and VRAM vs. perplexity curves.
- **Emotional Angle:** The satisfaction of decoding cryptic Hugging Face and Ollama filenames into a clear mental dashboard you can control.
- **Visual Sub-Blocks:** Decoding the Alphabet Soup of Model Tags (Translates intimidating quantization nomenclature into a simple numerical storage concept.); What 'Q4_K_M' Actually Means Letter by Letter (Deconstructs the exact technical meaning of `Q`, `K`, and `S/M/L` suffixes in llama.cpp/Ollama.); Why Coding Models Feel Quantization More Than Chat Models (Contrasts quantization sensitivity in code generation versus general prose.); Pulling Specific Quantization Tags in Ollama (Shows exact Ollama tag syntax for pulling specific GGUF quantization tiers.)
- **Connected Chapters:** /blog/24-billion-parameters-versus-my-8gb-vram-hardware-wall, /blog/why-bigger-ai-models-arent-always-better-speed-vs-size, /blog/why-100k-context-windows-became-my-new-ai-obsession, /blog/running-ai-models-on-my-own-hardware-the-local-llm-rabbit-hole
### BLOG #24: What 5–8 Tokens Per Second Actually Feels Like: The Human Experience of Waiting on Local AI
- **URL:** https://naborajs.dev/blog/what-5-to-8-tokens-per-second-feels-like-in-local-ai
- **Category:** AI & Engineering | **Timeline:** 2025–2026 (Local LLM Human-Factor Experiments)
- **Summary:** When a ~15 GB 24B model spills across VRAM and system memory and crawls at 5 to 8 tokens per second, you can literally watch the model think word by word—and you quickly learn why throughput matters. — Adapting prompt engineering specifically for low-throughput local models: banning conversational filler, demanding diff-only code blocks, and writing sharper initial prompts. — Understanding both the meditative appeal of watching a giant model reason locally and the exact threshold where cloud GPU offloading becomes mandatory.
- **Documented Facts:** Measured sustained generation speeds of 5 to 8 tokens per second when running ~15 GB 24B parameter models across split RTX 5060 VRAM and DDR5 system memory. | Calculated that polite AI boilerplate ('Sure! Here is the updated code...') wastes 8–12 seconds of wall-clock time per turn at 6 tokens/second. | Developed concise system prompts that cut response token length by 60%+ to make 24B local inference usable.
- **Technical Angle:** Prefill (prompt evaluation) vs. decode (token generation) latency, token-to-word ratios in code vs. prose, and concise diff-based system prompting.
- **Emotional Angle:** The strange mix of awe and impatience watching a 24-billion-parameter brain materialize code character by character on a laptop screen at midnight.
- **Visual Sub-Blocks:** Staring at the Blinking Cursor at 1:00 AM (Captures the phenomenology of low-throughput local inference.); Why 6 Tokens/Sec Feels Okay for Prose—and Brutal for Code (Explains the structural difference between prose tokenization and code tokenization under high latency.); The 'Zero-Fluff Diff' System Prompt I Built for 5–8 tok/s Models (Shows how prompt engineering can compensate for hardware memory bandwidth limits.); The Unexpected Benefit: Slow AI Makes You Write Better Prompts (Reflects on how high latency enforces deliberate engineering communication.)
- **Connected Chapters:** /blog/24-billion-parameters-versus-my-8gb-vram-hardware-wall, /blog/why-bigger-ai-models-arent-always-better-speed-vs-size, /blog/building-my-own-ai-endpoint-ollama-openai-v1-and-cloud-gpus, /blog/my-search-for-the-right-coding-ai-devstral-qwen-gemma-dolphin, /blog/why-100k-context-windows-became-my-new-ai-obsession
### BLOG #25: Why 100K+ Context Windows Became My New Obsession (and the Hidden Cost of Context)
- **URL:** https://naborajs.dev/blog/why-100k-context-windows-became-my-new-ai-obsession
- **Category:** AI & Engineering | **Timeline:** 2025–2026 (Long-Context LLM Experiments)
- **Summary:** Feeding an entire multi-file codebase into a 100K+ token context window feels like magic—right up until the KV cache explodes your VRAM allocation before the first token even streams. — Discovering how the Key-Value (KV) cache pre-allocates memory proportional to context length, layer count, and attention heads—and using GQA, Flash Attention, and KV cache quantization (`q8_0`/`q4_0`) to tame it. — Designing disciplined context-packing pipelines Locally (16K–32K) and offloading true 100K+ full-repo context sessions to ~30 GB cloud GPUs.
- **Documented Facts:** Investigated why Ollama models silently forgot top-of-prompt files when using the default `num_ctx 2048` / `4096` setting. | Tested raising `num_ctx` toward 32K, 64K, and 100K+ tokens on long-context models (including Qwen 2.5 Coder and Devstral-Small-2 24B). | Observed KV cache VRAM consumption scaling up by multiple gigabytes at 64K–128K context, forcing layers out of 8 GB VRAM unless Flash Attention and KV quantization are enabled.
- **Technical Angle:** Grouped-Query Attention (GQA), Key-Value (KV) cache memory calculation, `OLLAMA_FLASH_ATTENTION`, `OLLAMA_KV_CACHE_TYPE`, and Ollama `num_ctx` pre-allocation.
- **Emotional Angle:** The thrill of pasting an entire multi-file feature into a single prompt, followed by the shock of watching VRAM spike to 100% before a single word is generated.
- **Visual Sub-Blocks:** The Mystery of the Model With Amnesia (Exposes a critical real-world pitfall in local LLM tooling: silent context truncation at default `num_ctx`.); What the KV Cache Actually Is (and Why It Eats VRAM) (Explains the mechanics of Key-Value caching in autoregressive transformers.); Unlocking Larger Contexts With Flash Attention and Quantized KV Cache (Provides the exact environment variables (`OLLAMA_FLASH_ATTENTION`, `OLLAMA_KV_CACHE_TYPE`) to halve KV cache memory usage.); 32K Local Context vs. 100K+ Cloud GPU Context (Shows why 100K+ context windows were the catalyst for moving beyond local-only hardware.)
- **Connected Chapters:** /blog/quantization-explained-q4-q5-q6-q8-in-real-vram-tests, /blog/24-billion-parameters-versus-my-8gb-vram-hardware-wall, /blog/my-search-for-the-right-coding-ai-devstral-qwen-gemma-dolphin, /blog/building-my-own-ai-endpoint-ollama-openai-v1-and-cloud-gpus
### BLOG #26: Building My Own AI Endpoint: Connecting Ollama, Dual-T4 Cloud GPUs, and OpenAI-Compatible /v1 APIs
- **URL:** https://naborajs.dev/blog/building-my-own-ai-endpoint-ollama-openai-v1-and-cloud-gpus
- **Category:** AI & Engineering | **Timeline:** 2025–2026 (Hybrid Cloud-Local AI Infrastructure)
- **Summary:** When 8 GB of laptop VRAM wasn't enough for heavy models and huge context windows, I started experimenting with ~30 GB dual-T4 cloud GPU setups and exposing Ollama through an OpenAI-compatible `/v1` API. — Deploying Ollama across dual NVIDIA Tesla T4 GPUs (2x 15–16 GB = ~30–32 GB combined VRAM), tensor-splitting 24B+ models across both cards, and exposing the standardized `/v1/chat/completions` endpoint. — Turning my laptop into a lightweight cockpit that talks seamlessly to a 30 GB cloud GPU brain using standard OpenAI SDKs and IDE extensions.
- **Documented Facts:** Experimented with ~30 GB dual-T4 cloud GPU environments to overcome the 8 GB VRAM limit of the local RTX 5060 laptop. | Configured Ollama to distribute 24B+ parameter models across both T4 GPUs so 100% of weights and KV cache reside in dedicated VRAM. | Integrated Ollama's built-in OpenAI-compatible `/v1` endpoint with local TypeScript scripts, coding tools, and agent workflows.
- **Technical Angle:** Multi-GPU tensor/layer splitting across dual 16 GB T4 cards (`CUDA_VISIBLE_DEVICES=0,1`), Ollama `/v1` OpenAI API specification, streaming SSE chunks, and hybrid local-remote routing.
- **Emotional Angle:** The rush of seeing a 15 GB 24B model—which crawled at 6 tok/s on my laptop—load 100% into 30 GB of dual-T4 VRAM and stream answers effortlessly while my laptop fans stayed dead silent.
- **Visual Sub-Blocks:** Breaking Free From the 8 GB Ceiling (Marks the architectural transition from purely local inference to self-managed cloud GPU endpoints.); How Dual-T4 Multi-GPU Splitting Works in Ollama (Explains how layer parallelism across two 15 GB T4 GPUs solves both the weight and KV cache limits.); Pointing the Standard OpenAI SDK at My Custom Ollama /v1 Endpoint (Demonstrates drop-in OpenAI `/v1` SDK compatibility between local Ollama and remote cloud GPU instances.); Commercial Per-Token APIs vs. Self-Hosted Cloud GPU Endpoint (Compares managed commercial LLM APIs against self-hosted cloud GPU inference.)
- **Connected Chapters:** /blog/what-cloudflare-tunnels-taught-me-about-networking-and-apis, /blog/24-billion-parameters-versus-my-8gb-vram-hardware-wall, /blog/running-ai-models-on-my-own-hardware-the-local-llm-rabbit-hole, /blog/thinking-about-cloud-gpu-servers-during-exam-season, /blog/why-100k-context-windows-became-my-new-ai-obsession
### BLOG #27: What Cloudflare Tunnels and NSGO Taught Me About Networking, APIs, and Infrastructure
- **URL:** https://naborajs.dev/blog/what-cloudflare-tunnels-taught-me-about-networking-and-apis
- **Category:** AI & Engineering | **Timeline:** 2025–2026 (Networking, Tunnels & NSGO Experiments)
- **Summary:** Exposing a local or cloud Ollama server over HTTPS through a Cloudflare Tunnel—and experimenting with projects like NSGO—taught me more about how packets, headers, and endpoints work than any textbook diagram. — Using `cloudflared` outbound tunnels to bridge `localhost:11434` to an HTTPS edge URL without opening inbound ports, while solving streaming timeouts, `OLLAMA_ORIGINS`, and auth headers in my NSGO experiments. — Gaining deep, practical fluency in HTTP/2, Server-Sent Events (SSE), reverse proxies, and secure API gateway design—skills that directly powered EDITH and my multi-agent systems.
- **Documented Facts:** Used Cloudflare Tunnels (`cloudflared`) to expose local and cloud-hosted Ollama `/v1` endpoints over HTTPS without opening inbound router or VM firewall ports. | Built and iterated on the `NSGO` project namespace while testing custom API routing, client integrations, and tunnel endpoints. | Debugged real-world HTTP issues including `Host` header rewriting, `OLLAMA_ORIGINS` CORS policies, and Server-Sent Events (SSE) token streaming over reverse proxies.
- **Technical Angle:** Outbound-only `cloudflared` tunnels vs. NAT port forwarding, HTTP `Host` header validation, `OLLAMA_HOST` / `OLLAMA_ORIGINS`, SSE (`text/event-stream`) chunked transfer encoding, and zero-trust bearer authentication.
- **Emotional Angle:** The eureka moment when a request sent from a browser or phone app travels across Cloudflare's edge, down an encrypted tunnel, triggers a GPU, and streams tokens back in real time.
- **Visual Sub-Blocks:** Why `localhost:11434` Was an Island (Frames the real-world networking problem that every self-hoster encounters when moving beyond localhost.); How an Outbound Cloudflare Tunnel Flips Networking Upside Down (Explains the security and NAT-traversal mechanics of outbound reverse tunnels.); The Three Bugs Every Tunnel Builder Hits (and How I Fixed Them) (Documents three specific, high-signal infrastructure fixes: `OLLAMA_ORIGINS`, `--http-host-header`, and SSE streaming keep-alive.); Why Never Leaving an Unauthenticated AI Tunnel Open Matters (Reinforces secure engineering practices when exposing self-hosted AI APIs.)
- **Connected Chapters:** /blog/building-my-own-ai-endpoint-ollama-openai-v1-and-cloud-gpus, /blog/why-a-teenager-who-lives-on-the-internet-cares-about-privacy, /blog/why-i-keep-renaming-my-projects-ns-gaming-devspace-codex-nsgo, /blog/the-idea-behind-edith-whatsapp-ai-agent-by-ns, /blog/24-billion-parameters-versus-my-8gb-vram-hardware-wall
### BLOG #28: My Search for the Right Coding AI: Testing Devstral-Small-2 24B, Qwen Coder, Gemma, and Dolphin 8B
- **URL:** https://naborajs.dev/blog/my-search-for-the-right-coding-ai-devstral-qwen-gemma-dolphin
- **Category:** AI & Engineering | **Timeline:** 2025–2026 (Coding LLM Bake-Off & Model Routing)
- **Summary:** Not all coding models think the same way: testing `devstral-small-2:24b`, Qwen coding variants, Gemma models, and `dolphin3:8b` side by side showed me the difference between benchmark scores and real repo edits. — Discovering each model family's distinct 'personality': Devstral 24B's methodical repo-scale agentic planning, Qwen Coder's surgical syntax mastery, Gemma's clean structured reasoning, and Dolphin 8B's direct, zero-lecture instruction following. — Building a multi-model coding toolkit where each model is assigned to the exact engineering role it performs best.
- **Documented Facts:** Benchmarked `devstral-small-2:24b`, Qwen coding models, Gemma models, and `dolphin3:8b` on real full-stack TypeScript, Node.js, and Python tasks. | Observed `devstral-small-2:24b` excelling at multi-step repository navigation and tool-use formatting despite running at 5–8 tok/s locally (or full speed on dual-T4 cloud GPUs). | Used `dolphin3:8b` and Qwen 7B/8B variants for low-latency, in-VRAM coding loops where zero refusal friction and fast token streaming were top priorities.
- **Technical Angle:** Agentic tool-calling schemas, Fill-in-the-Middle (FIM) code completion, instruction-tuning datasets, and multi-model task routing.
- **Emotional Angle:** The fascination of realizing that open-weight neural networks trained by different labs (Mistral, Alibaba Qwen, Google Gemma, Eric Hartford's Dolphin) genuinely have distinct coding habits and personalities.
- **Visual Sub-Blocks:** Why I Stopped Trusting Synthetic Coding Leaderboards (Contrasts synthetic single-function benchmarks (HumanEval) against multi-file repository editing.); The Four Contenders: How Each Model Actually Thinks (Summarizes empirical strengths of Devstral-Small-2 24B, Qwen Coder, Gemma, and Dolphin 3 8B.); Agentic Repo Editing (Devstral 24B) vs. High-Speed Inline Coding (Qwen / Dolphin 8B) (Shows how task complexity and hardware constraints dictate model routing.); My Multi-Model Router Config for Coding Tasks (Codifies the multi-model cascade and task-routing pattern in TypeScript.)
- **Connected Chapters:** /blog/why-bigger-ai-models-arent-always-better-speed-vs-size, /blog/what-5-to-8-tokens-per-second-feels-like-in-local-ai, /blog/why-i-explore-open-and-uncensored-ai-models-technical-curiosity, /blog/24-billion-parameters-versus-my-8gb-vram-hardware-wall
### BLOG #29: Why I Experiment With Open and Unfiltered Models: Curiosity About Alignment and Refusal Layers
- **URL:** https://naborajs.dev/blog/why-i-explore-open-and-uncensored-ai-models-technical-curiosity
- **Category:** AI & Engineering | **Timeline:** 2025–2026 (AI Alignment & Open-Weights Research)
- **Summary:** When you test open-weights and unfiltered variants like Dolphin or uncensored Qwen builds in a local sandbox, the real fascination isn't breaking rules—it's seeing how fine-tuning and refusal guardrails actually shape a model's behavior. — Comparing base/instruct models against dataset-filtered (Dolphin) and refusal-abliterated variants locally to observe how RLHF/DPO alignment layers sit on top of raw world knowledge. — Applying those exact insights in reverse when engineering EDITH: building deterministic, domain-specific business guardrails instead of relying on vague model morality.
- **Documented Facts:** Tested open-weight unfiltered models including `dolphin3:8b` and uncensored Qwen builds locally inside Ollama. | Compared how standard safety-aligned models vs. unfiltered models respond to benign cybersecurity, networking, and system-administration prompts. | Used insights from studying model refusal behavior to design explicit, deterministic business guardrails and human-handoff rules for the EDITH WhatsApp AI agent.
- **Technical Angle:** Pre-training vs. Supervised Fine-Tuning (SFT) vs. DPO/RLHF alignment, dataset curation (removing 'As an AI language model' refusals), directional representation engineering (abliteration), and system-prompt adherence.
- **Emotional Angle:** The mindset of a hardware/software tinkerer opening up the hood of an engine—curious about how the governor works, not to drive recklessly, but to understand the machine completely.
- **Visual Sub-Blocks:** The Day an AI Lectured Me About My Own Local Server (Shows a concrete, legitimate developer frustration (false-positive refusals during security testing) that motivates exploring unfiltered models.); The Three Layers Inside Every Modern LLM (Explains the technical difference between dataset-filtered uncensored models (like Dolphin) and standard RLHF models.); Why 'Unfiltered' Matters for System Prompt Authority (Distinguishes between baked-in model refusals and developer-controlled system prompt alignment.); The Irony: Studying Unfiltered Models Made Me Better at Building Guardrails (Connects local curiosity about unfiltered models directly to responsible production AI engineering in Blog 30 and beyond.)
- **Connected Chapters:** /blog/my-search-for-the-right-coding-ai-devstral-qwen-gemma-dolphin, /blog/running-ai-models-on-my-own-hardware-the-local-llm-rabbit-hole, /blog/teaching-an-ai-business-agent-not-to-hallucinate-human-handoff, /blog/why-a-teenager-who-lives-on-the-internet-cares-about-privacy, /blog/the-idea-behind-edith-whatsapp-ai-agent-by-ns
### BLOG #30: The Idea Behind EDITH: My Vision for the WhatsApp AI Agent by NS
- **URL:** https://naborajs.dev/blog/the-idea-behind-edith-whatsapp-ai-agent-by-ns
- **Category:** Profit & Business | **Timeline:** 2025–2026 (The Birth & Evolution of EDITH)
- **Summary:** What started as a simple prototype concept for a tea business evolved into my most ambitious software architecture yet: EDITH, a multi-account WhatsApp AI agent built to handle real business operations. — Designing EDITH not as a toy Q&A chatbot, but as an operational WhatsApp business system combining verified product catalogs, lead CRM pipelines, order drafting, anti-hallucination guardrails, and a 10-module control dashboard. — Turning months of local AI and infrastructure experiments into a cohesive flagship product architecture: 'EDITH — WhatsApp AI Agent by NS'.
- **Documented Facts:** Originated EDITH from an initial prototype concept designed around automating product inquiries, pricing, and order flows for a tea business. | Architected EDITH ('WhatsApp AI Agent — by NS') as a multi-account WhatsApp business operations platform rather than a simple website chat widget. | Designed core subsystems covering product catalog grounding, lead capture/qualification, order management, anti-hallucination escalation, and a 10-module operator dashboard.
- **Technical Angle:** Multi-tenant WhatsApp session/webhook orchestration, grounded catalog retrieval (zero-hallucination pricing), structured state transitions (Lead -> Inquiry -> Quote -> Order), and human-in-the-loop escalation.
- **Emotional Angle:** The pride of watching a simple local prototype inspired by the tea trade around Siliguri grow into a serious, multi-module software system bearing the signature 'by NS'.
- **Visual Sub-Blocks:** Where It Started: A Tea Business Prototype in Siliguri (Documents the authentic origin of EDITH as a tea business prototype that evolved into a full multi-industry WhatsApp AI platform.); Why Generic Website Chatbots Fail Where WhatsApp Agents Win (Explains the strategic product positioning of EDITH around WhatsApp commerce rather than web widgets.); The Four Pillars of the EDITH Vision (Outlines the four core architectural pillars of the EDITH platform.); From Tea SKU Inquiry to Structured Order Intent (Shows the TypeScript domain schema connecting multi-account WhatsApp sessions, catalog guardrails, and the 10 dashboard modules.)
- **Connected Chapters:** /blog/i-didnt-want-a-chatbot-i-wanted-an-ai-workflow-agent, /blog/teaching-an-ai-business-agent-not-to-hallucinate-human-handoff, /blog/building-ai-around-whatsapp-leads-catalogs-pricing-and-orders, /blog/the-dashboard-behind-edith-10-modules-for-business-ai, /blog/when-two-ai-systems-talk-friday-and-edith-agent-orchestration, /blog/24-billion-parameters-versus-my-8gb-vram-hardware-wall, /blog/quantization-explained-q4-q5-q6-q8-in-real-vram-tests, /blog/building-my-own-ai-endpoint-ollama-openai-v1-and-cloud-gpus, /blog/why-100k-context-windows-became-my-new-ai-obsession, /blog/what-cloudflare-tunnels-taught-me-about-networking-and-apis, /blog/why-i-explore-open-and-uncensored-ai-models-technical-curiosity
### BLOG #31: I Didn't Want a Chatbot. I Wanted an Agent That Executes Real Workflows.
- **URL:** https://naborajs.dev/blog/i-didnt-want-a-chatbot-i-wanted-an-ai-workflow-agent
- **Category:** AI Systems & Agent Architecture | **Timeline:** 2025–2026 • EDITH Agent Architecture & Workflow Prototyping
- **Summary:** A chatbot replies with paragraphs of text and hopes you go away. An AI agent reads structured tables, checks pricing rules, updates lead states, and knows when to call a human. — By separating conversational phrasing from deterministic state transitions—treating the LLM as an intent router and tool caller around SQL/CSV records—EDITH shifted from a toy chatbot into a workflow engine. — I learned that real software engineering in AI isn't writing longer system prompts; it's building the database schemas, state guards, and execution pipelines around the model.
- **Documented Facts:** Naboraj Sarkar (Nishant / NS GAMING) architected EDITH as a WhatsApp business workflow agent prototype rather than an open-ended chatbot. | The EDITH architecture processes structured business entities including leads, product catalogs, SQL/CSV pricing rules, orders, and human handoff states. | The workflow separates natural-language intent parsing from deterministic state mutations and price calculations.
- **Technical Angle:** Event-driven state machines, typed JSON tool-calling schemas, SQL/CSV catalog hydration, and strict separation of LLM parsing from arithmetic execution.
- **Emotional Angle:** Intellectual curiosity and builder discipline—refusing to settle for a flashy demo that fails under real-world operational constraints.
- **Visual Sub-Blocks:** Text Generator vs. Deterministic Workflow Agent (Distinguishes decorative conversational AI from operational workflow automation.); The 5-Step Message-to-Action Execution Pipeline (Demonstrates how sandwiching an LLM between deterministic pre-checks and post-commit validations creates reliability.); Typed Intent & Workflow State Transition Contract (Enforces a strict JSON schema between the LLM inference step and the WhatsApp dispatch layer.); Never Let a Neural Network Guess Arithmetic or Policy (Assigns probabilistic tasks to AI and deterministic tasks to classic software engineering.)
- **Connected Chapters:** /blog/the-idea-behind-edith-whatsapp-ai-agent-by-ns, /blog/teaching-an-ai-business-agent-not-to-hallucinate-human-handoff, /blog/building-ai-around-whatsapp-leads-catalogs-pricing-and-orders, /blog/when-two-ai-systems-talk-friday-and-edith-agent-orchestration, /blog/the-dashboard-behind-edith-10-modules-for-business-ai
### BLOG #32: Teaching an AI Business Agent Not to Hallucinate: Grounded Data and Human Handoff
- **URL:** https://naborajs.dev/blog/teaching-an-ai-business-agent-not-to-hallucinate-human-handoff
- **Category:** AI Systems & Agent Architecture | **Timeline:** 2025–2026 • EDITH Grounding & Human-in-the-Loop Engineering
- **Summary:** I wasn't trying to build a giant corporation—I just wanted my WhatsApp business agent to stop making up prices and instead escalate to a human owner whenever it didn't know the answer. — I designed a strict three-part guardrail in EDITH: closed-world grounding against verified SQL/CSV records + Knowledge Rack entries, explicit confidence & policy checks, and an instant Human Handoff state switch. — An AI that knows when to say 'Let me connect you with the owner right now' is infinitely more trustworthy than an AI that pretends to know everything.
- **Documented Facts:** EDITH prevents hallucinated business responses by grounding answers strictly in catalog records, SQL/CSV pricing rules, and Knowledge Rack entries. | When a customer inquiry falls outside grounded data or exceeds pricing rule boundaries, EDITH triggers a Human Handoff state. | The Human Handoff module mutes automated replies for that contact and surfaces the conversation inside the operator dashboard.
- **Technical Angle:** Closed-world retrieval constraints, deterministic confidence/discount boundary checks, and session-level state locks (`AI_ACTIVE` vs `HUMAN_HANDOFF`).
- **Emotional Angle:** Pragmatic humility and respect for real-world consequences—building software that protects trust rather than showing off.
- **Visual Sub-Blocks:** The 3 Pillars of Closed-World Business Grounding (Constrains the LLM to a closed world of verified business data rather than open-ended pre-training weights.); Deterministic Guardrail & Human Handoff Escalation Gate (Replaces vague prompt instructions with hard, testable if-statements that force human escalation when boundaries are crossed.); Anatomy of a Clean Human Handoff Lifecycle (Prevents the classic 'zombie bot' bug where an automated responder interrupts a live human conversation.); Prompt-Level 'Please Don't Lie' vs. System-Level Grounding (Shows why reliable AI engineering treats hallucinations as a systems architecture problem rather than a prompt-wording problem.)
- **Connected Chapters:** /blog/the-idea-behind-edith-whatsapp-ai-agent-by-ns, /blog/i-didnt-want-a-chatbot-i-wanted-an-ai-workflow-agent, /blog/building-ai-around-whatsapp-leads-catalogs-pricing-and-orders, /blog/the-dashboard-behind-edith-10-modules-for-business-ai
### BLOG #33: Building AI Around WhatsApp: Leads, Catalogs, Pricing Rules, Orders, and Anti-Spam State
- **URL:** https://naborajs.dev/blog/building-ai-around-whatsapp-leads-catalogs-pricing-and-orders
- **Category:** SaaS & Product Architecture | **Timeline:** 2025–2026 • EDITH Commerce Data Modeling & Anti-Spam Logic
- **Summary:** In India, business happens on WhatsApp—which means an effective AI agent has to understand catalogs, SQL/CSV pricing rules, last-message timestamps, and how to follow up without spamming customers. — I modeled five interlocking tables—Leads, Catalog, Pricing Rules (SQL/CSV), Orders, and Anti-Spam Cooldown State—so every message respects timing rules while moving commercial state forward. — Designing around real Indian commerce patterns taught me how database schema design directly shapes both customer experience and account safety.
- **Documented Facts:** EDITH models WhatsApp business operations around leads, product catalogs, SQL/CSV pricing rules, orders, and anti-spam state. | Pricing rules in EDITH support structured SQL/CSV lookups so quantity tiers and discount boundaries are evaluated deterministically. | Anti-spam state tracking records message timestamps and follow-up counts to prevent repetitive automated messaging.
- **Technical Angle:** Relational schema design for leads/orders, SQL/CSV tiered price lookup queries, webhook burst debouncing, and timestamp-based cooldown state machines.
- **Emotional Angle:** Empathy for both the busy shop owner and the customer—engineering software that feels calm, accurate, and respectful.
- **Visual Sub-Blocks:** The 5 Interlocking Data Tables of EDITH (Transforms unstructured WhatsApp chat streams into normalized relational state.); Burst Debouncing & Anti-Spam Follow-Up Guard (Enforces strict mathematical boundaries on automated follow-ups so the agent never spams a contact.); Bulk-Blast Spam Bots vs. State-Aware WhatsApp Workflow Agent (Contrasts spammy broadcast scripts with stateful, customer-respecting conversational commerce.); From Fragmented Chat to Structured Order Draft (Shows the end-to-end transformation of informal messaging into structured ERP-ready order rows.)
- **Connected Chapters:** /blog/the-idea-behind-edith-whatsapp-ai-agent-by-ns, /blog/teaching-an-ai-business-agent-not-to-hallucinate-human-handoff, /blog/the-dashboard-behind-edith-10-modules-for-business-ai, /blog/can-a-16-year-old-student-build-a-real-saas-product, /blog/i-didnt-want-a-chatbot-i-wanted-an-ai-workflow-agent
### BLOG #34: The Dashboard Behind the AI: Designing Overview, Live Chat, Leads, Catalog, Knowledge Rack & Handoff
- **URL:** https://naborajs.dev/blog/the-dashboard-behind-edith-10-modules-for-business-ai
- **Category:** SaaS & Product Architecture | **Timeline:** 2025–2026 • EDITH 10-Module Operator Cockpit & UI Architecture
- **Summary:** An AI agent running in the background is invisible until you build its cockpit: 10 modular dashboard views—from Overview and Live Chat to Pricing Rules, Knowledge Rack, and Human Handoff. — I architected a 10-module web cockpit where every database table and agent state has a dedicated, transparent control surface—from Knowledge Rack cards to a one-click Human Handoff takeover toggle. — Building the 10 modules proved to me that great AI UX is really about observability and human control.
- **Documented Facts:** The EDITH dashboard architecture comprises 10 modules: Overview, Live Chat, Leads, Catalog, Pricing Rules, Knowledge Rack, Orders, Human Handoff, Analytics, and Settings. | The Knowledge Rack module stores verified business policies as structured records rather than raw prompt strings. | The Human Handoff and Live Chat modules allow a human operator to inspect AI decisions and take over conversations.
- **Technical Angle:** Modular React dashboard routing, citation badges linking LLM outputs to Knowledge Rack/Pricing IDs, and real-time session takeover state toggles.
- **Emotional Angle:** Craftsmanship and clarity—taking pride in turning complex backend tables into a clean, calm cockpit.
- **Visual Sub-Blocks:** The 4 Functional Zones of the 10-Module Cockpit (Organizes 10 operational views into four intuitive mental models for a business operator.); Editing Raw System Prompts vs. Editing the Knowledge Rack (Separates developer prompt engineering from operator business-rule management.); Tracing an Escalation Across 4 Dashboard Modules (Illustrates how the 10 modules act as interconnected lenses on the same underlying state machine.); Dashboard Module Registry & Route Configuration (Defines the strongly typed structure of the 10-module EDITH operator interface.)
- **Connected Chapters:** /blog/the-idea-behind-edith-whatsapp-ai-agent-by-ns, /blog/why-i-love-building-dashboards-turning-chaos-into-control, /blog/building-ai-around-whatsapp-leads-catalogs-pricing-and-orders, /blog/teaching-an-ai-business-agent-not-to-hallucinate-human-handoff
### BLOG #35: When Two AI Systems Need to Talk to Each Other: Exploring the FRIDAY + EDITH Concept
- **URL:** https://naborajs.dev/blog/when-two-ai-systems-talk-friday-and-edith-agent-orchestration
- **Category:** AI Systems & Agent Architecture | **Timeline:** 2025–2026 • FRIDAY + EDITH Multi-Agent Architecture Exploration
- **Summary:** What happens when one AI agent handles customer conversations while a second coordinating agent monitors workflows, notifications, and project tasks? That question led me to the FRIDAY + EDITH architecture concept. — I split the architecture into two distinct roles communicating over a typed event bus: EDITH as the external commercial front-line agent, and FRIDAY as the internal operator's chief-of-staff coordinator. — Designing the boundary between FRIDAY and EDITH taught me the core principle of multi-agent engineering: strict role isolation and structured inter-agent events.
- **Documented Facts:** Naboraj Sarkar (Nishant) explored the FRIDAY + EDITH architecture concept to study how two specialized AI systems coordinate workflows. | In this concept, EDITH handles external WhatsApp customer workflows (leads, catalogs, pricing, handoffs) while FRIDAY focuses on internal operator coordination, notifications, and task monitoring. | The architecture emphasizes structured event-driven communication and role separation rather than monolithic agent design.
- **Technical Angle:** Multi-agent role isolation, prompt-injection sandboxing, typed JSON event envelopes, and priority-ranked human-in-the-loop escalation queues.
- **Emotional Angle:** Imaginative yet grounded—showing how a 16-year-old builder turns creative inspiration into structured systems thinking.
- **Visual Sub-Blocks:** Role Isolation: E.D.I.T.H. vs. F.R.I.D.A.Y. (Enforces the principle of least privilege across multi-agent boundaries.); The Inter-Agent Typed Event Bus Protocol (Eliminates multi-agent token drift and infinite conversational loops by enforcing typed event contracts.); End-to-End Dual-Agent Escalation & Resolution Loop (Demonstrates asynchronous, human-in-the-loop multi-agent coordination without losing state.); Why Multi-Agent Systems Should Communicate Like Microservices (Applies classic distributed-systems discipline to multi-agent AI design.)
- **Connected Chapters:** /blog/the-idea-behind-edith-whatsapp-ai-agent-by-ns, /blog/i-didnt-want-a-chatbot-i-wanted-an-ai-workflow-agent, /blog/the-day-i-started-taking-ai-seriously-beyond-chatbots, /blog/where-i-want-to-go-next-ai-agents-saas-and-creator-systems, /blog/the-dashboard-behind-edith-10-modules-for-business-ai
### BLOG #36: Can a Teenager Build a SaaS Product? Thinking About Multi-Tenant Software While in Class 10
- **URL:** https://naborajs.dev/blog/can-a-16-year-old-student-build-a-real-saas-product
- **Category:** SaaS & Product Architecture | **Timeline:** 2025–2026 • Multi-Tenant SaaS Architecture & Prototype Engineering
- **Summary:** Designing per-business database isolation, multi-session WhatsApp routing, and React + Python + Supabase architectures while studying for Class 10 exams is a strange but addictive balancing act. — By redesigning my prototype around tenant-scoped database rows (`tenant_id` + Row-Level Security concepts), isolated session workers, and a clean React + Python + Supabase stack, I learned the difference between a single-user script and a SaaS platform. — Even before launching commercially, architecting for multi-tenancy at 16 gave me a deep foundation in production software engineering.
- **Documented Facts:** Naboraj Sarkar (Nishant), born 19 August 2010 in Siliguri, West Bengal, designed multi-tenant SaaS architectures while studying in Class 10. | His SaaS architecture for EDITH explores per-business database isolation, multi-session WhatsApp routing, and a React + Python + Supabase stack. | The work is framed honestly as architectural engineering and developing software prototypes rather than commercial revenue claims.
- **Technical Angle:** Multi-tenant PostgreSQL/Supabase schema design (`tenant_id` + RLS concepts), stateless Python/Node webhook workers, and channel-to-tenant routing.
- **Emotional Angle:** Grounded ambition—combining teenage curiosity with professional-grade architectural standards and zero fake hype.
- **Visual Sub-Blocks:** Single-Tenant Hobby Script vs. Multi-Tenant SaaS Architecture (Highlights the architectural leap from single-user scripts to multi-tenant SaaS systems.); Tenant-Scoped Webhook Context Resolver (Prevents cross-tenant data leakage by resolving and locking the tenant context at the outermost request boundary.); The 4-Layer Stack Behind My SaaS Prototypes (Shows how modern open-source and cloud-native tools let a solo student architect full-stack multi-tenant systems.); Honest Framing: Architecture First, Hype Never (Grounds the narrative in intellectual honesty, separating genuine architectural skill from internet hype.)
- **Connected Chapters:** /blog/why-i-keep-thinking-about-business-from-cool-code-to-useful-product, /blog/the-idea-behind-edith-whatsapp-ai-agent-by-ns, /blog/my-class-10-life-in-one-sentence-study-build-repeat, /blog/when-a-website-becomes-more-than-a-website-platform-architecture
### BLOG #37: Why I Keep Thinking About Business: Moving From 'Cool Project' to 'Useful Product'
- **URL:** https://naborajs.dev/blog/why-i-keep-thinking-about-business-from-cool-code-to-useful-product
- **Category:** SaaS & Product Architecture | **Timeline:** 2025–2026 • Transitioning from Hobby Code to Product Thinking
- **Summary:** Writing code just to see if it compiles is fun, but thinking about whether a real business owner would rely on your software forces you to grow up as an engineer. — When I started evaluating my code through the eyes of a busy shop owner—asking 'Does this save time, prevent mistakes, or protect leads?'—my engineering priorities flipped toward reliability, simplicity, and guardrails. — Business thinking didn't make coding less creative for me; it gave my code a purpose.
- **Documented Facts:** Naboraj Sarkar (Nishant) focuses his software engineering practice on moving from experimental coding scripts to structured, useful product architectures. | In his EDITH prototype, business constraints directly drove features like deterministic SQL/CSV pricing lookups, anti-spam cooldowns, and the 10-module operator cockpit. | Every operational exception in the architecture is designed to surface an actionable alert or human handoff rather than failing silently.
- **Technical Angle:** Operational error recovery, human-centric dashboard alerts, deterministic data grounding, and defensive input handling.
- **Emotional Angle:** Thoughtful maturity and builder pride—finding deeper satisfaction in reliability and usefulness than in superficial complexity.
- **Visual Sub-Blocks:** Developer-Ego Features vs. Operator-Utility Features (Contrasts technology-first vanity engineering with problem-first product engineering.); My 4-Question 'Useful Product' Filter Before Writing Code (Provides a repeatable product-engineering checklist for evaluating technical features.); Business Constraints Make You a Better Programmer (Explains how real-world product constraints elevate raw technical execution.); Designing Errors for Humans, Not Just Stack Traces (Translates low-level runtime exceptions into actionable operator workflows.)
- **Connected Chapters:** /blog/can-a-16-year-old-student-build-a-real-saas-product, /blog/the-idea-behind-edith-whatsapp-ai-agent-by-ns, /blog/building-ai-around-whatsapp-leads-catalogs-pricing-and-orders, /blog/where-i-want-to-go-next-ai-agents-saas-and-creator-systems, /blog/teaching-an-ai-business-agent-not-to-hallucinate-human-handoff
### BLOG #38: My Class 10 Life in One Sentence: Study, Build, Repeat
- **URL:** https://naborajs.dev/blog/my-class-10-life-in-one-sentence-study-build-repeat
- **Category:** Student & Builder Life | **Timeline:** 2026 • Class 10 CBSE Academic Year & Evening Builder Sessions, Siliguri
- **Summary:** By day it's CBSE Mathematics, Physics, Chemistry, and Social Science maps; by evening it's GitHub commits, Ollama terminals, and UI components—and both sides are teaching me how to think. — Instead of treating schoolwork and coding as enemies fighting over the same clock, I realized they train complementary mental muscles: Mathematics and Physics sharpen step-by-step rigor, while coding turns abstract logic into living systems. — Three words—Study, Build, Repeat—became the anchor that keeps me grounded as a student in Siliguri and growing as an engineer.
- **Documented Facts:** Naboraj Sarkar (online identity Nishant Sarkar / NS GAMING) is a Class 10 CBSE student from Siliguri, West Bengal, born on 19 August 2010. | His daily routine balances Class 10 academic subjects (Mathematics, Physics, Chemistry, Social Science) with evening software development (GitHub, Ollama, UI/UX, and AI prototypes). | He structures his evening builder sessions into incremental milestones so academic responsibilities remain prioritized.
- **Technical Angle:** Cognitive cross-training between formal mathematical/physical problem-solving and incremental software engineering workflows.
- **Emotional Angle:** Earnest, disciplined, and relatable—capturing both the excitement of evening coding and the real responsibility of Class 10 academics.
- **Visual Sub-Blocks:** Anatomy of a 'Study, Build, Repeat' Weekday in Siliguri (Shows a realistic, sustainable time-blocking structure for a Class 10 student builder.); How CBSE Class 10 Subjects Map to Software Engineering (Connects Indian Class 10 curriculum topics directly to core computer science mental models.); Small Commits Beat All-Nighters During the School Year (Explains why academic time constraints actually encourage professional incremental development habits.); What Lives on My Desk in Siliguri (2026 Snapshot) (Grounds the narrative in concrete, authentic artifacts from Nishant's daily workspace.)
- **Connected Chapters:** /blog/who-is-naboraj-sarkar-nishant-student-gamer-builder, /blog/why-i-sometimes-know-the-syllabus-but-still-dont-feel-ready, /blog/when-your-passion-becomes-your-distraction-coding-during-study-hours, /blog/thinking-about-cloud-gpu-servers-during-exam-season
### BLOG #39: Why I Sometimes Know the Syllabus but Still Don't Feel Exam-Ready
- **URL:** https://naborajs.dev/blog/why-i-sometimes-know-the-syllabus-but-still-dont-feel-ready
- **Category:** Study Philosophy & Systems | **Timeline:** 2026 • Class 10 CBSE Syllabus Progression & Exam Reflection
- **Summary:** Finishing chapters like Real Numbers, Polynomials, Linear Equations, Quadratic Equations, and Arithmetic Progressions feels great—until you look at Trigonometry, Surface Areas, Statistics, and Probability waiting ahead. — I realized that 'knowing the concept' and 'being exam-ready' are two different states—just like code that compiles vs. code that survives load testing—and mapped out a structured revision loop to bridge the gap. — Admitting that I don't always feel 100% ready took away the panic and replaced it with a clear, honest checklist of what to practice next.
- **Documented Facts:** In his Class 10 Mathematics progression, Naboraj Sarkar completed chapters including Real Numbers, Polynomials, Pair of Linear Equations, Quadratic Equations, and Arithmetic Progressions while preparing for Trigonometry, Surface Areas and Volumes, Statistics, and Probability. | He reflects candidly on the psychological gap between knowing syllabus concepts and feeling fully exam-ready under timed conditions. | He uses structured error-isolation and spaced practice loops to convert conceptual familiarity into timed exam readiness.
- **Technical Angle:** Applying software QA and load-testing mental models (bug classification, closed-book reproduction, spaced regression testing) to CBSE Mathematics preparation.
- **Emotional Angle:** Vulnerable, honest, and constructive—acknowledging real student nervousness without exaggeration or defeatism.
- **Visual Sub-Blocks:** 'I Understand the Chapter' vs. 'I Am Exam-Ready' (Pinpoints the cognitive difference between passive comprehension and active, timed recall.); My Class 10 Mathematics Syllabus Map: Base vs. Frontier (Documents the real CBSE Class 10 Mathematics progression from completed foundational algebra to upcoming high-drill chapters.); My 4-Step 'Debugging Loop' for Chapters That Don't Feel Ready (Turns vague exam self-doubt into a deterministic 4-step error-correction loop.); Why Not Feeling 100% Ready Is Actually a Healthy Signal (Reframes academic self-doubt as a useful diagnostic signal rather than a source of panic.)
- **Connected Chapters:** /blog/my-class-10-life-in-one-sentence-study-build-repeat, /blog/when-your-passion-becomes-your-distraction-coding-during-study-hours, /blog/what-school-teaches-that-self-taught-coding-doesnt, /blog/the-grayscale-study-app-i-wanted-to-build-for-myself
### BLOG #40: Algorithms vs. Homework: Navigating Instagram, Telegram Communities, and Attention as a Student Creator
- **URL:** https://naborajs.dev/blog/instagram-telegram-and-attention-social-media-as-creator-and-student
- **Category:** Creator Strategy & Focus | **Timeline:** 2023–2026 • Creator Distribution Across YouTube, Instagram, Discord & Telegram
- **Summary:** When you use Instagram, YouTube, Discord, and Telegram communities to build an audience, social media isn't just entertainment—it's both your distribution engine and your biggest focus trap. — I drew a hard architectural line between 'Creator Producer Mode' (publishing, community updates, intentional networking) and 'Passive Consumer Mode' (algorithmic scrolling), pairing scheduled batches with strict notification boundaries. — Learning how attention algorithms work from the inside as a creator gave me the exact blueprint I needed to defend my own focus as a student.
- **Documented Facts:** Naboraj Sarkar (Nishant / NS GAMING) uses Instagram, YouTube, Discord, and Telegram communities as part of his creator and builder ecosystem. | He distinguishes between algorithmic discovery platforms (Instagram/YouTube) and direct community channels (Telegram/Discord) while managing attention boundaries during Class 10 studies. | He applies a 'Producer Firewall' workflow—muting real-time group pings and batching social updates after completing daily study and coding blocks.
- **Technical Angle:** Push-vs-pull notification architecture, cognitive context-switching costs, algorithmic feed mechanics vs. broadcast community graphs.
- **Emotional Angle:** Self-aware, balanced, and disciplined—refusing both mindless scrolling and anti-technology cynicism.
- **Visual Sub-Blocks:** Algorithmic Discovery Platforms vs. Community Messaging Hubs (Categorizes creator platforms into Discovery Engines vs. Community Hubs alongside their specific cognitive traps.); The 4-Rule 'Producer Firewall' for Student Creators (Establishes a practical behavioral firewall between content creation and academic/engineering deep work.); What Being a Creator Taught Me About Visual & Social Engineering (Shows how creator experience compounds into design, communication, and self-regulation skills.); Don't Let Your Distribution Engine Eat Your Production Engine (Frames the balance between deep work (production) and social media (distribution) in systems terms.)
- **Connected Chapters:** /blog/when-your-passion-becomes-your-distraction-coding-during-study-hours, /blog/the-grayscale-study-app-i-wanted-to-build-for-myself, /blog/subscriber-counts-are-weird-vanity-metrics-vs-real-community, /blog/starting-over-when-everyone-thinks-you-already-made-it
### BLOG #41: When Your Biggest Passion Becomes Your Biggest Distraction: Thinking About AI Models While Studying
- **URL:** https://naborajs.dev/blog/when-your-passion-becomes-your-distraction-coding-during-study-hours
- **Category:** Focus & Student Builder Life | **Timeline:** 2025–2026 (Class 10 Study & Builder Balance)
- **Summary:** Most advice about student distraction assumes you're wasting time on mindless scrolling, but what do you do when the thing pulling you away from textbook revision is an idea for a new AI architecture? — Building a 'Capture-and-Park' scratchpad protocol that honors the technical idea in 45 seconds without opening VS Code or breaking the study block. — Protecting academic discipline while preserving my best engineering insights for dedicated post-study build windows.
- **Documented Facts:** Naboraj Sarkar (Nishant) balances Class 10 CBSE board preparation in Siliguri with active software and AI experimentation. | Experienced recurring cognitive switching where ideas for local LLMs, AI agents, and web architectures surfaced during textbook study sessions. | This friction directly inspired concepts for focus tooling, including a grayscale study application with configurable friction.
- **Technical Angle:** Analyzes human working memory as a single-threaded processor with high context-switching latency and introduces an analog interrupt queue.
- **Emotional Angle:** The tug-of-war between excitement over a fresh technical breakthrough and the quiet guilt of an unfinished textbook chapter.
- **Visual Sub-Blocks:** The Chemistry Notebook Margin Full of JSON Schemas (Productive passions bypass standard willpower defenses because they disguise themselves as meaningful work.); Cheap Dopamine vs. Builder Dopamine During Study Hours (Distinguishing between passive scrolling and active creative hijacking is essential for young builders.); My 45-Second 'Capture-and-Park' Protocol (Externalizing an idea onto paper frees working memory without triggering screen-based context switching.); Modelling Cognitive Context Switching as a State Machine (Treating human focus like a single-threaded CPU with expensive context-switch overhead makes discipline logical.)
- **Connected Chapters:** /blog/my-class-10-life-in-one-sentence-study-build-repeat, /blog/the-grayscale-study-app-i-wanted-to-build-for-myself, /blog/instagram-telegram-and-attention-social-media-as-creator-and-student, /blog/thinking-about-cloud-gpu-servers-during-exam-season
### BLOG #42: The Study App I Wanted to Build for Myself: Grayscale UI and Configurable Friction
- **URL:** https://naborajs.dev/blog/the-grayscale-study-app-i-wanted-to-build-for-myself
- **Category:** PKM & Focus Systems | **Timeline:** 2025–2026 (Focus Systems & Mobile Architecture Concept)
- **Summary:** Instead of extreme lock-down blockers that you immediately uninstall, I started designing a concept for a study app in Flutter and Kotlin that turns high-dopamine feeds grayscale and lets you tune your own focus friction. — Designing 'Configurable Friction'—a multi-tier behavioral architecture using grayscale display filters, intentional delay gates, and study-mode profiles. — A blueprint for humane focus software that works with human psychology instead of triggering rebellion and uninstallation.
- **Documented Facts:** Naboraj Sarkar conceived a mobile study app architecture combining Flutter for UI and Kotlin for native Android system controls. | The core design centers on grayscale visual dampening and configurable friction rather than rigid app blocking. | This concept grew directly out of his experience managing dual-use study/social apps during Class 10 CBSE preparation.
- **Technical Angle:** Explores hybrid mobile architecture using Flutter for declarative monochrome UI and Kotlin `MethodChannel` bridges for Android package monitoring and visual overlays.
- **Emotional Angle:** Moving from frustration with rigid productivity apps to the creative satisfaction of designing a more empathetic solution.
- **Visual Sub-Blocks:** Why I Uninstalled Every 'Hard Lock' Focus App Within 72 Hours (Binary blocking fails when dual-use platforms host both educational utility and algorithmic distraction.); The 4 Tiers of Configurable Friction (Giving users agency over their friction level increases long-term adherence.); Flutter + Kotlin Bridge Concept for Focus Friction State (Combining cross-platform UI ergonomics with native OS capabilities is the cleanest way to build utility apps on Android.); Full-Color Dopamine UI vs. E-Ink Grayscale Study UI (Desaturating the interface lowers the variable reward loop without deleting functional utility.)
- **Connected Chapters:** /blog/when-your-passion-becomes-your-distraction-coding-during-study-hours, /blog/instagram-telegram-and-attention-social-media-as-creator-and-student, /blog/why-i-sometimes-know-the-syllabus-but-still-dont-feel-ready, /blog/why-i-started-coding-from-using-software-to-building-it
### BLOG #43: School Projects, But Make Them Look Professional: Why My Brain Treats A4 Assignments Like UI Design
- **URL:** https://naborajs.dev/blog/school-projects-but-make-them-look-like-a-product-launch
- **Category:** Design Systems & Academic Craft | **Timeline:** 2025–2026 (Class 10 Project Submissions)
- **Summary:** Even when an assignment is just a 12-to-20-page school project for English, Geography, Bengali, or Mathematics, I can't stop myself from thinking about cover typography, visual hierarchy, and print layout systems. — Applying web UI design systems—consistent spacing tokens, modular callout boxes, clear typographic scales, and diagram flows—to physical A4 pages. — Turning mandatory school coursework into high-craft editorial artifacts that are genuinely enjoyable to build and read.
- **Documented Facts:** As a Class 10 CBSE student in Siliguri, Naboraj Sarkar completed structured 12-to-20-page academic projects across subjects including English, Geography, Bengali, Mathematics, and History. | He systematically applied visual hierarchy, grid layouts, and editorial typography learned from web development and video editing to his school project submissions.
- **Technical Angle:** Translates CSS grid mental models, typographic scales, binding gutter math, and component callout boxes into A4 document architecture.
- **Emotional Angle:** Turning routine school obligations into creative design challenges so the work feels energizing rather than repetitive.
- **Visual Sub-Blocks:** Once You Learn UI Design, You Can Never Unsee Bad Page Margins (Craft isn't something you turn on only for code repositories—it becomes the default lens through which you approach every piece of work.); Default Copy-Paste Assignment vs. UI-Engineered Project File (Restraint and consistency beat decorative clutter every single time.); My 5-Point Pre-Print Checklist Across Subjects (A repeatable design system saves time while raising quality across every subject.); Thinking of an A4 Page Like CSS Grid Tokens (Print layout and web layout share the exact same mathematical foundations of proportion, rhythm, and contrast.)
- **Connected Chapters:** /blog/making-of-a-global-world-engineering-a-20-page-history-project, /blog/my-dream-of-making-a-cinematic-website-neon-ui-and-motion, /blog/building-my-personal-website-ui-seo-aeo-and-identity, /blog/my-class-10-life-in-one-sentence-study-build-repeat
### BLOG #44: From Silk Routes to Information Networks: Structuring My 20-Page Project on The Making of a Global World
- **URL:** https://naborajs.dev/blog/making-of-a-global-world-engineering-a-20-page-history-project
- **Category:** Knowledge Systems & History | **Timeline:** 2025–2026 (Class 10 History Project Deep Dive)
- **Summary:** Working on a 15-to-20-page History project on the Pre-Modern World—covering the Silk Routes, cultural exchange, and the Columbian Exchange—made me realize that ancient trade networks were the original global internet. — Mapping pre-modern globalization using the exact same network concepts I learned from studying APIs, routing nodes, bandwidth, and payloads. — A cohesive 20-page historical dossier where silk, spices, silver, crops, faiths, and pathogens are understood as packets moving across a global mesh network.
- **Documented Facts:** Naboraj Sarkar developed a 15-to-20-page Class 10 CBSE History project on 'The Making of a Global World', focusing on the Pre-Modern World. | The project covered the Silk Routes linking Asia with Europe and North Africa, food travels and the Columbian Exchange (potatoes, maize, tomatoes, chillies), and 16th-century conquest, disease, and trade. | He structured the historical narrative using systems-thinking diagrams, commodity-flow tables, and editorial page layouts.
- **Technical Angle:** Maps historical globalization to network concepts: multi-hop routing (Silk Route hubs), bi-directional payload transfer (silk/spices vs. silver), dependency fragility (Irish Potato Famine SPOF), and biological zero-day vulnerabilities (smallpox in isolated populations).
- **Emotional Angle:** Intellectual curiosity—the excitement of realizing that a school History chapter and modern computer networking tell the exact same story at different speeds.
- **Visual Sub-Blocks:** When Ancient Caravans Looked Like Packet Routing (Systems thinking lets you translate concepts between computer networking and human history.); How I Architected the 20-Page History Dossier (Breaking a 20-page research project into modular chapters prevents repetition and builds a clear argument.); The Silk Routes vs. The Modern Internet: A Systems Comparison (Analogical mapping anchors new humanities knowledge onto existing technical schemas.); The Potato Lesson: How Single-Dependency Architecture Creates Fragility (History teaches the exact same resilience lessons that distributed systems engineering teaches.)
- **Connected Chapters:** /blog/school-projects-but-make-them-look-like-a-product-launch, /blog/what-school-teaches-that-self-taught-coding-doesnt, /blog/my-class-10-life-in-one-sentence-study-build-repeat, /blog/what-cloudflare-tunnels-taught-me-about-networking-and-apis
### BLOG #45: What School Teaches That Coding Doesn't: Structure, Fundamentals, and Finishing Hard Things
- **URL:** https://naborajs.dev/blog/what-school-teaches-that-self-taught-coding-doesnt
- **Category:** Philosophy & Education | **Timeline:** 2025–2026 (Reflections on Formal Schooling vs. Self-Taught Tech)
- **Summary:** Self-taught developers love to say you can learn everything online, but being in Class 10 taught me something side projects never do: how to finish work even when you didn't choose the syllabus. — Realizing that formal school exams and fixed syllabi train 'non-negotiable completion'—the exact muscle needed to ship real production software. — Respecting both worlds: using self-taught coding for speed and creativity, and school structure for endurance, breadth, and discipline.
- **Documented Facts:** Naboraj Sarkar is a Class 10 CBSE student in Siliguri who taught himself video editing, web development, and local AI orchestration alongside his formal schooling. | He credits structured school coursework with teaching completion discipline, breadth across subjects, and unaided problem-solving.
- **Technical Angle:** Examines the difference between novelty-driven side-project loops and full-scope specification compliance without external search or AI assistance.
- **Emotional Angle:** Gratitude and maturity—recognizing the value in difficult routines instead of complaining about them.
- **Visual Sub-Blocks:** The Graveyard of Half-Finished GitHub Folders (Unconstrained freedom trains exploration, but external constraints train endurance.); Self-Taught Sandbox Learning vs. Structured Academic Training (Combining autonomous curiosity with structured fundamentals produces a far stronger thinker than either alone.); How School Subjects Quietly Upgraded My Engineering Brain (Foundational subjects build the mental hardware that programming languages run on.); The Rule of the Last 20% (Grit is measured in the unglamorous middle and final 20% of an obligation.)
- **Connected Chapters:** /blog/my-class-10-life-in-one-sentence-study-build-repeat, /blog/why-i-sometimes-know-the-syllabus-but-still-dont-feel-ready, /blog/the-first-time-coding-stopped-feeling-like-homework, /blog/making-of-a-global-world-engineering-a-20-page-history-project
### BLOG #46: The Strange Life of a Teenager Who Thinks About Servers and Cloud GPUs During Exam Season
- **URL:** https://naborajs.dev/blog/thinking-about-cloud-gpu-servers-during-exam-season
- **Category:** Nishant's Journey & Dual Worlds | **Timeline:** 2025–2026 (Exam Season & Cloud GPU Experiments)
- **Summary:** There is something genuinely funny about memorizing chemical equations or geography map points while a tiny part of your brain is calculating whether dual NVIDIA T4 GPUs can fit a larger context window. — Accepting the duality with humor instead of stress, and scheduling cloud notebook runs strictly as post-revision rewards. — A grounded, memorable snapshot of what it actually feels like to grow up as a student builder in India during the AI wave.
- **Documented Facts:** Naboraj Sarkar (Nishant) experimented with both local 8GB RTX 4060 GPU inference and cloud dual-T4 GPU notebooks while preparing for Class 10 CBSE examinations in Siliguri. | He used time-boxed evening windows and automated bash/notebook scripts to test AI endpoints without derailing his exam revision schedule.
- **Technical Angle:** Contrasts 8GB local VRAM constraints against 32GB (2×16GB) cloud T4 VRAM budgets for running 14B–24B GGUF models with extended KV-cache contexts.
- **Emotional Angle:** Self-aware humor mixed with quiet pride in managing two demanding worlds at sixteen.
- **Visual Sub-Blocks:** Alluvial Soil on Page 8, Tensor Parallelism in My Head (Growing up as a builder in the middle of the AI revolution creates hilarious everyday contrasts.); The Mental Math That Randomly Pops Up While Studying (Hands-on hardware constraints make memory arithmetic second nature.); My Two Daily Dashboards During Exam Season (Compartmentalizing physical and digital workspaces keeps both identities healthy.); The 30-Second Health Check Script for Nightly Cloud Sessions (Automating setup boilerplate protects small, time-boxed coding windows during busy academic months.)
- **Connected Chapters:** /blog/my-class-10-life-in-one-sentence-study-build-repeat, /blog/building-my-own-ai-endpoint-ollama-openai-v1-and-cloud-gpus, /blog/when-your-passion-becomes-your-distraction-coding-during-study-hours, /blog/who-is-naboraj-sarkar-nishant-student-gamer-builder
### BLOG #47: What My Failed Experiments Taught Me: Slow Models, Renamed Repos, Lost Channels, and Rebuilding
- **URL:** https://naborajs.dev/blog/what-my-failed-experiments-taught-me-models-repos-and-rebuilds
- **Category:** Retrospective & Resilience | **Timeline:** 2022–2026 (Multi-Year Post-Mortem Across Media & Code)
- **Summary:** Looking back, I didn't learn the most from the things that worked on the first try—I learned from the 50K channel I lost, the 24B model that crawled at 5 tokens per second, and the architectures I had to rewrite. — Conducting an honest post-mortem across four years of experiments—treating every setback as empirical data that upgraded my judgment. — Realizing that my identity isn't tied to any single channel, model, or repository, because the real asset is the ability to rebuild from zero.
- **Documented Facts:** Naboraj Sarkar (Nishant) grew his original NS GAMING YouTube channel to ~50,000 subscribers before losing it to a security compromise. | He benchmarked a 24B parameter local LLM (`cognitivecomputations_Dolphin-Mistral-24B-Venice-Edition-Q4_K_M.gguf`, ~14.34 GB) on his 8GB RTX 4060 + 16GB RAM PC, measuring ~5.26 tokens/sec (~19 seconds per 100 tokens). | He iteratively evolved and renamed his technical projects across multiple generations (`NS GAMING`, `DevSpace`, `Codex`, `NSGO`, `EDITH`).
- **Technical Angle:** Synthesizes lessons across operational security, VRAM/PCIe memory-bandwidth bottlenecks in local LLM inference, and iterative software refactoring.
- **Emotional Angle:** Hard-earned resilience—turning past disappointments into quiet confidence.
- **Visual Sub-Blocks:** Why Portfolios Should Include a 'Failures & Post-Mortems' Tab (Success confirms what you already know; failure forces you to learn what you were missing.); What I Thought Would Happen vs. What Actually Happened (Documenting the gap between expectation and reality builds senior-level engineering judgment early.); My 4-Step Post-Mortem Loop When an Experiment Breaks (A structured post-mortem turns emotional frustration into reusable documentation.); The Real Asset They Can't Hack or Delete (Skill and resilience are the only truly non-custodial assets a creator or engineer owns.)
- **Connected Chapters:** /blog/what-losing-a-50k-subscriber-youtube-channel-taught-me, /blog/24-billion-parameters-versus-my-8gb-vram-hardware-wall, /blog/why-i-keep-renaming-my-projects-ns-gaming-devspace-codex-nsgo, /blog/the-story-isnt-finished-building-in-public-at-16
### BLOG #48: Why a Teenager Who Loves the Internet Also Cares About Privacy and Owning His Own Systems
- **URL:** https://naborajs.dev/blog/why-a-teenager-who-lives-on-the-internet-cares-about-privacy
- **Category:** Privacy, Security & Self-Hosting | **Timeline:** 2024–2026 (Identity Boundaries & Local-First Systems)
- **Summary:** Being a public creator on YouTube and GitHub doesn't mean putting your entire private life on display—understanding how servers, APIs, and data flows work makes you care deeply about boundaries and digital control. — Architecting a clear boundary layer—sharing code, lessons, and creative work openly as Nishant / Naboraj Sarkar while keeping personal environments, local AI memory, and private life strictly protected. — Sustainable public creation backed by local-first AI inference, environment-variable hygiene, and respectful personal boundaries.
- **Documented Facts:** Naboraj Sarkar operates publicly under his real name and creator persona Nishant / NS GAMING (`@Nishant_sarkar`) while maintaining strict boundaries around private offline life. | His technical stack emphasizes local LLM execution via Ollama, self-managed API tunnels, and owned web platforms.
- **Technical Angle:** Contrasts third-party cloud API data custody against `localhost` LLM inference, secret redaction pipelines, and scoped Cloudflare tunnel exposure.
- **Emotional Angle:** Calm confidence and responsibility—valuing both public creative expression and a quiet, protected personal life.
- **Visual Sub-Blocks:** When You Inspect Network Payloads, You Stop Treating the Cloud as 'Magic' (Technical literacy naturally turns casual internet users into privacy-conscious architects.); Building in Public vs. Oversharing in Public (You can be 100% authentic about your ideas, struggles, and architectures without exposing private coordinates or other people's privacy.); Sanitizing Payloads and Enforcing Local-First Routing (Privacy is strongest when it is enforced at the code and router layer rather than left to manual memory.); My 4 Rules for Teen Builders Operating on the Public Internet (Good operational security lets young builders participate in the global tech ecosystem safely.)
- **Connected Chapters:** /blog/why-i-became-nishant-online-identity-vs-creator-branding, /blog/what-cloudflare-tunnels-taught-me-about-networking-and-apis, /blog/running-ai-models-on-my-own-hardware-the-local-llm-rabbit-hole, /blog/who-is-naboraj-sarkar-nishant-student-gamer-builder
### BLOG #49: Where I Want to Go Next: Autonomous AI Agents, Multi-Tenant SaaS, and High-Craft Content
- **URL:** https://naborajs.dev/blog/where-i-want-to-go-next-ai-agents-saas-and-creator-systems
- **Category:** Future Roadmap & Vision | **Timeline:** 2026 & Beyond (Next-Horizon Builder Roadmap)
- **Summary:** Everything I've experimented with so far—NS GAMING, 4K editing, React & Supabase platforms, local LLMs, and EDITH—feels like the foundation for what I want to build over the next few years. — Seeing how all three pillars—Autonomous AI Systems, Multi-Tenant SaaS Engineering, and High-Craft Visual Storytelling—compound into one rare builder profile. — A realistic, step-by-step horizon roadmap grounded in fundamentals rather than hype.
- **Documented Facts:** Naboraj Sarkar (Nishant) has built prototypes spanning 4K video editing (NS GAMING), full-stack web applications (React, TypeScript, Supabase), local/cloud LLM pipelines (Ollama), and multi-agent architectures (EDITH and FRIDAY). | His forward roadmap focuses on combining autonomous agent orchestration, multi-tenant SaaS engineering, and high-craft creator systems.
- **Technical Angle:** Outlines next-generation requirements for deterministic agent tool-calling, hybrid local/cloud LLM routing, and multi-tenant Supabase Row-Level Security (RLS) architectures.
- **Emotional Angle:** Focused anticipation—feeling grounded in what has been built so far while staying hungry for deeper engineering mastery.
- **Visual Sub-Blocks:** Connecting the Dots Forward From a Desk in Siliguri (Multidisciplinary skills that look scattered at 13 become a rare competitive advantage by 16 and 20.); The Three Technical Horizons I'm Building Toward (Clear technical milestones turn long-term ambition into daily learning targets.); Where My Systems Are Today vs. Where I Want Them Next (Self-awareness about current technical ceilings defines the next syllabus.); A Blueprint for Hybrid Agent Routing (Next-Gen EDITH Concept) (The future of practical AI agents isn't calling the biggest model for everything—it is smart routing across deterministic code, small local models, and deep reasoning endpoints.)
- **Connected Chapters:** /blog/the-idea-behind-edith-whatsapp-ai-agent-by-ns, /blog/when-two-ai-systems-talk-friday-and-edith-agent-orchestration, /blog/can-a-16-year-old-student-build-a-real-saas-product, /blog/the-story-isnt-finished-building-in-public-at-16
### BLOG #50: The Story Isn't Finished: Building in Public at 16 While Still Figuring Myself Out
- **URL:** https://naborajs.dev/blog/the-story-isnt-finished-building-in-public-at-16
- **Category:** Series Finale & Manifesto | **Timeline:** 2010–2026+ (16 Years Old — End of Volume 1, Beginning of What's Next)
- **Summary:** Maybe the point of documenting all 50 chapters—from Siliguri classrooms and Free Fire matches to losing a 50K channel, testing 24B models, and architecting EDITH—isn't to prove I've arrived, but to show what learning by building looks like in real time. — Writing and connecting all 50 chapters at sixteen—not as a monument to a finished career, but as a living knowledge graph of how curiosity compounds year by year. — Closing Volume 1 with gratitude, humility, and momentum: Class 10 student by day, builder by night, and just getting started.
- **Documented Facts:** Naboraj Sarkar (born 19 August 2010, from Siliguri, West Bengal, India) is a 16-year-old Class 10 CBSE student, creator (Nishant / NS GAMING, `@Nishant_sarkar`), and software/AI builder. | Across this 50-blog series, he documented his evolution from mobile gaming and growing a ~50K YouTube channel to 4K editing, full-stack web engineering, local/cloud LLM benchmarking, and multi-agent architectures (EDITH and FRIDAY).
- **Technical Angle:** Summarizes the end-to-end knowledge graph architecture—bidirectional wiki-links, structured metadata, visual explainers, and full-stack + AI systems synthesis.
- **Emotional Angle:** Deep gratitude, grounded self-awareness, and the quiet excitement of knowing that sixteen is just the beginning.
- **Visual Sub-Blocks:** Why Write 50 Blogs at Sixteen Instead of Waiting Until I'm 'Successful'? (Documenting the journey in real time is more useful to other learners than waiting to write a polished victory lap years later.); Who I Was at the Start of This Arc vs. Who I Am at Blog 50 (Tools change every eighteen months; the transition from consumer to systems thinker lasts a lifetime.); Volume 1 Complete: Initializing the Next Loop (A milestone is a checkpoint to reflect and refactor, never an excuse to stop learning.); A Note to Anyone Building From a Small Room With Limited Hardware (The ultimate purpose of sharing this 50-blog graph is to prove that curiosity and persistence beat waiting for perfect conditions.)
- **Connected Chapters:** /blog/who-is-naboraj-sarkar-nishant-student-gamer-builder, /blog/what-losing-a-50k-subscriber-youtube-channel-taught-me, /blog/the-idea-behind-edith-whatsapp-ai-agent-by-ns, /blog/what-my-failed-experiments-taught-me-models-repos-and-rebuilds, /blog/where-i-want-to-go-next-ai-agents-saas-and-creator-systems