Skip to content

00  AI agents Private models Native apps Building since 2010

NitinJaiswal

Founder @ AdultCoder

Building a whole ecosystem of enterprise AI coding tools: simple to use, with privacy and compliance designed in, for software where both are critical.

Nitin Jaiswal
AdultCoder
  1. todo_writeTask list saved: 0 of 8 done. In progress: Scaffold Astro project + install.
  2. write_filewrote 3569 bytes to src/components/home/Proof.astro
  3. edit_fileapplied 2 edit operation(s) to src/components/home/Hero.astro
  4. bash[build] 39 page(s) built in 753ms
  5. Compacted the conversation: 82 earlier messages, about 217K tokens, are now a summary, and the work goes on from it.
  6. Tasks
    • [x] Scaffold Astro project + install
    • [x] Subtle sound: opt-in toggle, ambient pad, UI sounds
    • [x] Forbidden-string scan script over src + dist, fix hits
    • [x] Final QA: Chrome + WebKit, both themes, 390 and 1440
  7. git_commit[main d86756f] Keep the hero's notes on screen and give the glass its blur back
Real lines from the sessions: AdultCoder built most of this site, and my personal one.
  • Building AdultCoder, full time
  • Small models, distilled and fine-tuned
  • A native AI harness, written in Rust
  • Shipping software since 2010

01Why

We wanted the speed of AI coding without giving up our code, our keys or our team’s knowledge.

AdultCoder’s founding line

  1. 01

    Fifteen years of shipping at scale

    Products used by millions, in fintech, media and e-commerce. They taught me one thing: speed without control doesn’t ship.

  2. 02

    AI-first, already at work

    In my last role I gave talks on MCP, pushed for AI coding tools in the team’s everyday work, and architected Figma-to-code generation driven by design variables.

  3. 03

    Agents that start fast and finish nothing

    AI writes the code. Teams still need the plan, the checks, the receipts and the privacy.

  4. 04

    So I built the one I needed

    AdultCoder: private by design, built for adults shipping software, not prompt tourists.

02AdultCoder

Pro-grade AI coding. Your data never leaves your control.

Version 0.2: a terminal shell and a Mac app. Beta soon.

Every run

Six things it does, so the work arrives finished and accounted for.

Plans first

It plans the work, and builds once you approve.

Sandboxed

Every command runs in a sandbox.

Brakes

Auto mode, with brakes: it stops before anything destructive.

Your keys stay yours

The agent never sees them.

Receipts

An audit log that can’t be quietly edited: one shared record.

Budgets

Set the budget. Get the finished work.

Proof, not promises

It built most of this site. And my personal one.

Real sessions, start to finish: it read and wrote the files, checked the pages in a browser, compacted its own context when the conversation grew long, and committed the work.

Visit my personal site nitinjaiswal.com

AdultCoder at work in a terminal on my personal site: file reads, writes and checks scroll past, the conversation is compacted, and the task list shows five of eleven steps done
AdultCoder at work on my personal site · the folder names are blurred
To start
0.01 s
Peak memory for a turn
56 MB
A full turn in a Rust repository of
200,000 lines

A native engine, not a web app in disguise · measured for adultcoder.tech

03Native

A native app, drawn on my own UI toolkit.

A native agent chat, written in Rust on a UI toolkit of my own and drawn on the GPU: a proof of concept, ahead of AdultCoder’s native desktop app.

01 / 06
  1. 01A calm startAsk for anything, or begin from a suggestion.
  2. 02At workA request goes in, and the agent gets going.
  3. 03StreamingThe reply streams in, its code coloured as it arrives.
  4. 04FinishedA complete answer, ready to copy.
  5. 05Pictures inlineScreenshots, charts and files sit right in the conversation.
  6. 06Full-window viewerOpen one large; the rest wait as thumbnails.
The native agent chat, dark: the welcome screen: “What shall we work on?” with three suggested prompts
The native agent chat, light: the welcome screen: “What shall we work on?” with three suggested prompts
The native agent chat, dark: a request to write a Rust function, and the agent thinking
The native agent chat, light: a request to write a Rust function, and the agent thinking
The native agent chat, dark: the answer streaming in, its Rust code coloured as it arrives
The native agent chat, light: the answer streaming in, its Rust code coloured as it arrives
The native agent chat, dark: the finished answer: a Rust function with a usage example, and a note on how it works
The native agent chat, light: the finished answer: a Rust function with a usage example, and a note on how it works
The native agent chat, dark: a conversation about a blog post, with a grid of screenshots shown inline
The native agent chat, light: a conversation about a blog post, with a grid of screenshots shown inline
The native agent chat, dark: a screenshot opened full-window in the viewer, with the others as thumbnails
The native agent chat, light: a screenshot opened full-window in the viewer, with the others as thumbnails

The app’s own rendered screens · a scripted agent answers for now

A native agent chat in light mode: a conversation titled “Port the thumbnail worker from Python to Rust”, with highlighted Python code, two replies and a bar chart, beside a list of chats
The same native agent chat in dark mode, further into the conversation: a follow-up question, a reply with the chart, and the next answer arriving

Live, on my Mac

Light and dark, from the same code.

  • Drawn on the GPU, with its own layout engine
  • Nine design styles in six palettes, each held to accessible contrast, checked from the rendered pixels
  • Ready for screen readers
  • Long histories stay smooth
  • Clear states when the network goes, or a reply takes long
  • Light and dark, following the system
Lines of Rust in the toolkit, about
250,000
Tests, about
4,000
Since
June 2026

04Models

Small models that know your domain.

I distil small models from large ones and fine-tune them on domain-specific knowledge, then make them smaller and faster, with guardrails.

  1. 01DistilA large model teaches small ones the work that matters.
  2. 02Fine-tuneDomain-specific knowledge, so they know the field they work in.
  3. 03ShrinkSmaller and faster, to run on modest hardware.
  4. 04GuardGuardrails, so they stay on task.
Inside the GPU workstation: a GeForce RTX card under a liquid cooler reading 27 °C, lit by RGB fans
Where they’re trained, in the labLiquid-cooled

05Voice

A voice that stays on the device.

A real-time voice assistant that runs completely offline on a small device: it listens, talks back naturally in more than one language, copes with interruptions, and can look at a photo mid-conversation.

  • Listens
  • Talks back naturally, in more than one language
  • Copes with interruptions
  • Can look at a photo mid-conversation

Offline · on the device · no cloud in the loop

Coming, in time

Alongside this, I’m building something you’ll be able to hold: a small device that brings this work together. More when it is.

06How I build

I build with AI agents, AdultCoder included.

  • I set

    the direction, the rules and the bar.

  • Agents do

    the typing, in a sandbox, with brakes.

  • Checks decide

    what ships: tests, types, a browser, a scan.

AdultCoder, so far

Lines of Rust, about
250,000
Tests, about
2,400
Since
March 2026
  1. Before

    AI-first, already at work

    Talks on MCP, AI coding tools in the team’s everyday work, and Figma-to-code generation driven by design variables.

  2. February 2026

    Founded AdultCoder

    Full time, on private AI coding for enterprises.

  3. March 2026

    AdultCoder’s first lines

    A native engine, in Rust.

  4. June 2026

    The UI toolkit

    Drawn on the GPU, with its own layout engine, for native apps.

  5. November 2026

    All of it, together

    Small models distilled and fine-tuned on domain knowledge, the native agent chat, and AdultCoder’s launch site.

07The lab

Where it’s built.

A room of boards, screens and parts, a GPU in a box, robots on the bench, and a wall of the pioneers I look up to.

Boards, screens and a GPU in a box.

Robots on the bench.

And a wall of the pioneers I look up to.

The lab end to end: a keyboard, the desk and monitors under the wall of pinned notes
A servo-driven robotic arm on a table, wired to a controller
Camera modules and a servo driver board wired up for computer vision

08Before AdultCoder

Fifteen years of shipping, before AdultCoder.

Products used by millions, in fintech, media and e-commerce: staff, principal and architect roles, a company of my own, and a stage at the Topcoder Open finals.

Senior Staff Engineer

Forbes Advisor · 2024–2026

Already AI-first at work

  • Championed AI in engineering: gave talks inside the company on MCP (the Model Context Protocol), and pushed for AI coding tools in the team’s everyday work.
  • Architected Figma-to-code generation driven by the design system’s variables.
  • Led the frontend team while working on the backend, about 90% hands-on.
  • Revamped a highly complex, mission-critical, consumer-facing e-commerce product (B2C, live, at scale) with many journeys and 10+ third-party integrations, under strict security and compliance, with 95%+ test coverage and automation rules.
  • Worked closely with the compliance team on privacy and security.
    • Understanding and Prioritizing Data PrivacyLinkedIn · Dec 2025 · Privacy compliance
    • Learning GDPRLinkedIn · Dec 2025 · Privacy compliance
LimitPushCo-founder · 2023–2024

A spiritual social app with AI chat, a 3D game, and a team of twenty.

  • Built Ayodhya.ai (“Your Spiritual Ally”), a spiritual social app with an AI chat, in React Native while the team designed it, and a 3D game alongside. Neither launched; the demos are on video.
  • Wrote a custom video streaming engine for Ayodhya.ai, to keep its streaming costs down.
  • Learned After Effects and Premiere Pro together with the team, and made our promo videos ourselves, like Ayodhya.ai’s.
  • Grew the team from 0 to 20 across development, design, product, content and video.
  • Ran the company end to end alongside the product: hiring, payroll, legal and accounting.
Ayodhya.ai: Your Spiritual AllyYouTube · loads when played
Ayodhya.ai: the base version, demoYouTube · loads when played
A 3D game, work in progress: demoYouTube · loads when played

AI in the app, 2023–24

  • Mistral 7B, fine-tuned
  • A Rust inference pipeline for ComfyUI
  • ComfyUI video workflows
  • Agentic 3D design in Blender
  • Map segmentation
  • Robotic cars and arms
  • Songs, music and audio
  • After Effects and Premiere Pro
  • Inside the GPU workstation: a GeForce RTX card under a liquid cooler reading 27 °C, lit by RGB fans
    The RTX 4090 it all ran on, in the lab
  • ComfyUI, a sample screen from its project: a node graph that loads a model, encodes a prompt, samples, decodes and saves an image
    ComfyUI, where the app’s videos were madeA sample screen from the ComfyUI project (GPL-3.0); model names blurred
  • A Jupyter notebook, qlora_example: a fine-tuned model, NitinGPT, answering a YouTube comment about finding a job as a software engineer
    Fine-tuning with QLoRA, May 2024: “NitinGPT”, answering YouTube comments
  • A Jupyter notebook, stablediffusion_dreambooth: Stable Diffusion XL with its refiner, and the picture it made, an astronaut riding a unicorn past a rainbow
    Stable Diffusion XL and DreamBooth, in a notebook, May 2024
  • An agentic 3D design app: “create a modern Japanese style bedroom with blue jacuzzi” typed in a chat, and the bedroom built in 3D beside it
    An agentic 3D design app: a chat builds the room in Blender, with a local LLM on my GPU
  • The same app: “remove everything and create a human character from scratch again”, and a simple character built in the scene
    The same app, building a character from scratch
  • A Blender render of a room the app designed: a bed between two side tables, a blue rug, a chair and blue boxes
    One of its rooms, rendered in Blender
  • Aerial map segmentation: fields and trees outlined in green, buildings and roads in blue
    Map segmentation with SAM: fields and trees in green, buildings and roads in blue
  • The same aerial map, a second pass: more of the built-up land found, in blue
    A second pass, finding more of the built-up land

Robotics, in the lab

  • A two-wheeled rover chassis wired to a breadboard
    A smart car: a rover chassis, wired up
  • A servo-driven robotic arm on a table, wired to a controller
    A servo-driven robotic arm
  • Camera modules and a servo driver board wired up for computer vision
    Cameras and a servo driver, for computer vision
  • A sensor held in a clamp, wired to a microcontroller board
    A sensor, a clamp, a microcontroller
  • A Ludo game in React Native: an ornate board with a lotus in each corner, two players and the dice
    Ludo, in React Native · mine
  • MediCare, a part-time nursing jobs app in Flutter: the welcome screen: “Find part-time nursing jobs with ease and flexibility”
    MediCare, in Flutter · the team’s
  • MediCare, a part-time nursing jobs app in Flutter: signing in with a mobile number and a one-time code
    MediCare · signing in
  • MediCare, a part-time nursing jobs app in Flutter: the wallet: money in process and settled, and a transfer to the bank
    MediCare · the wallet
Principal Engineer · 2022–2023
  • Led the mobile app development revamp: rewriting Beem’s fintech super app (micro-loans, money transfer, tax filing), used by millions of people at scale, from scratch with the mobile and frontend team.
  • Solved the hard problems in the React Native app, among them a custom encryption layer for advanced security of sensitive data.
  • Integrated many fintech SDKs in the mobile app, among them Plaid, Socure, Stripe and Jscrambler.
  • Wrote the initial app architecture and design system, about 25% of the code, and helped hire the frontend team.

Reset and reinvent · 2021–2022

A year to learn: a social network MVP, and a token on Ethereum.

  • Stepped back during Covid: time with family, and a year of learning with no syllabus.
  • Went deep into React’s internals, and rethought cross-platform UX from first principles.
  • Built the MVP of Bharat Solves, a made-in-India social network app (“Together we progress”): a feed, travel, events and communities. It was never released.
  • To learn how tokens work, wrote an ERC-20 token, German Shepherd Dog (GSD), deployed it on Ethereum on 23 May 2021, and airdropped it to 20+ people.
GSD

German Shepherd Dog (GSD)

ERC-20 on Ethereum · 27 holders

Deployed 23 May 2021, 18:23 UTC, in block 12,492,182; airdropped from 30 May 2021 to 4 Jun 2021, in 25 transfers.

The contract on Etherscan Its creation

Transactions · latest 25From Etherscan’s public record
Transaction hashMethodBlockAge (UTC)From
0x71e8eb…17da04Transfer125675664 Jun 2021, 11:010x681f5F…0CdA
0x411e16…ee99b4Transfer125675554 Jun 2021, 10:590x681f5F…0CdA
0xfafcd2…a7e8a4Transfer125598563 Jun 2021, 06:160x681f5F…0CdA
0x2ce77f…138260Transfer125529462 Jun 2021, 04:440x681f5F…0CdA
0xacbee8…b8b072Transfer125529352 Jun 2021, 04:410x681f5F…0CdA
0x9ac671…1648a9Transfer125529262 Jun 2021, 04:390x681f5F…0CdA
All 25 transfers
0x2d3327…92528aTransfer125529232 Jun 2021, 04:380x681f5F…0CdA
0x44493f…093b47Transfer125529152 Jun 2021, 04:360x681f5F…0CdA
0x9adc97…bd024dTransfer1254194931 May 2021, 11:480x681f5F…0CdA
0x896d7c…39ca3fTransfer1254194931 May 2021, 11:480x681f5F…0CdA
0xdcb807…9016c2Transfer1254193431 May 2021, 11:440x681f5F…0CdA
0xb4fd55…fb979cTransfer1254193231 May 2021, 11:440x681f5F…0CdA
0x98d986…ed414aTransfer1254190731 May 2021, 11:390x681f5F…0CdA
0xbdc5aa…da4ed7Transfer1254189131 May 2021, 11:350x681f5F…0CdA
0xfa022e…091e3eTransfer1254187931 May 2021, 11:330x681f5F…0CdA
0x67d11d…a37167Transfer1253489330 May 2021, 09:270x681f5F…0CdA
0x0dab3e…ca8d8bTransfer1253485430 May 2021, 09:170x681f5F…0CdA
0x2637a3…d2541aTransfer1253484930 May 2021, 09:160x681f5F…0CdA
0x940956…2124bdTransfer1253474730 May 2021, 08:540x681f5F…0CdA
0x06d28c…05e942Transfer1253463030 May 2021, 08:270x681f5F…0CdA
0x9e41cd…090d0aTransfer1253462330 May 2021, 08:260x681f5F…0CdA
0xdc7bdd…de39eeTransfer1253451230 May 2021, 07:590x681f5F…0CdA
0x2cf3f9…44302aTransfer1253447530 May 2021, 07:510x681f5F…0CdA
0xae12d3…c6e567Transfer1253442230 May 2021, 07:400x681f5F…0CdA
0xbfcaa4…74cc1dTransfer1253440330 May 2021, 07:370x681f5F…0CdA
  • Bharat Solves’ feed, 2021: a post in English, Hindi and Kannada about the app, with its logo, “Together we progress”
    Bharat Solves · the feed
  • Bharat Solves, a social network app MVP, 2021: travel cards, events to book, and communities to join
    Bharat Solves · travel, events, communities
JavaScript Architect · 2018–2021

Pages I built at Al Jazeera, live today.

Built pages for aljazeera.com, aljazeera.net and Al Jazeera Balkans, and wrote several WordPress embeds.

One of the first engineers on the core revamp team, rewriting decade-old .NET systems with React, Node.js, a WordPress VIP editorial CMS and custom server-side rendering, for a global audience at scale.

Competitive engineer, Copilot and Reviewer · 2016–2018

Topcoder Open (TCO)

2 timefinalist

Topcoder’s world championship. I made the 2017 finals in two tracks, First2Finish and UI Prototype, and competed live in the US.

My Topcoder profile, happyesthete

Where it started · 2006–2010

Bachelor of Technology, Computer Science

Maharana Pratap Engineering College (MPEC), Kanpur

Affiliated to UPTU, the state’s technical university

  • A tiny operating system, from scratch

    In my final year, 2009–10, I built a tiny operating system from scratch with my team, based on the BrokenThorn OS Development Series.

  • UPTUhub.com, a social network for my college

    I built UPTUhub.com, a social network for my college: tutorials, solved exam papers, placement papers, a technical forum, blogs and jobs, in one place.

  • Embedded System & RoboticsBRICS, IIT Kanpur · 2009 · Training
UPTUhub.com in 2010: the home page of the social network I built for my college, with tutorials, solved papers, placement papers, forums, blogs and jobs
UPTUhub.com’s blogs in 2010: students’ posts, newest first

The full record

09Contact

Building something private with AI?Let’s talk.

Ideas by me, built with AdultCoder (opens adultcoder.tech in a new tab)