VVarun NidhiApplied AI & LLM Solutions Architect
Helping people make better calls under uncertainty

Varun Nidhi — I architect AI solutions
that turn messy operations into decisions teams trust.

I architect AI solutions for the hard end of real operations — anywhere the data is incomplete, the conditions are unforgiving, and a wrong call is expensive. I take it the whole way: discovery, architecture, model selection, integration, evaluation, rollout — and I lead the engineering teams that build it, keeping the boardroom and the engineers speaking the same language throughout.

Explore the toolsSee the case studies
Varun NidhiVarun Nidhi
B.Tech ECE · IIT Roorkee
Fifteen years in industrial tech since 2010 — from wireless to AI
Teams of ~20
Led engineering teams of around 20 across software, machine learning, and applied AI
Proven at scale
Engines that read 1.3M data points a second, analytics across billions of records, and reporting that went from weeks to hours
Across industries
Pipeline safety, defense, field mobility, education, finance, connected devices, and big data
01Financial Tools

Tools for making sense of markets — the same instinct, pointed at price.

TradingView Indicators

A growing library of Pine Script indicators for reading markets under uncertainty — published openly on TradingView for anyone to add to their charts.

Explore
Open Options Trader

A free, educational web app for exploring Indian F&O contracts — price options with Black-Scholes, browse expiry calendars, and build multi-leg strategies with payoff diagrams. Theoretical model prices, not live market data.

Open Options TraderExplore
Pigeon Station

A self-hosted station that turns financial news — RSS feeds and Telegram channels — into structured, filterable trading signals, with an LLM doing the reading so you're left with the call, not the noise.

Explore
02Creative Tools

Tools for making things — the same instinct, pointed at creative work.

Creator Lab

A self-contained creative-learning launcher — a Steam-style catalog of 64 curated tools and services across 15 creative categories, plus 8 guided learning paths that chain them into step-by-step projects like making your first short film or building a 2D game.

Creator LabMake
03Open Source

A few things I've open-sourced — free for anyone to pick up and run.

LLM-on-Web

An AI chat app that runs language models entirely in your browser — no server, no sign-up — with document Q&A built in.

View repoLive demo ↗
Prompt2Powerpoint

Turn a plain-language prompt into a finished PowerPoint deck, working with either a local model or a cloud API.

View repoLive demo ↗
QuickTag-Images

Point it at a folder of images and a local AI gives every file a sensible name and tags — organisation without the busywork.

View repoLive demo ↗
Browse github.com/van007
04Things I've Built

Products built to survive contact with the real world.

ShikshaOS

Education · LLM

An AI teaching assistant that turns a chapter of a school textbook into quizzes, exam papers, summaries and lesson plans — in the language the book is written in.

Showcase

ShikshaOS

Education · LLM

An AI teaching assistant that turns a chapter of a school textbook into quizzes, exam papers, summaries and lesson plans — in the language the book is written in.

Problem

Teachers spend their evenings writing the same materials over and over — quiz questions, an exam paper with proper sections and an answer key, a chapter condensed into notes. It's near-identical work happening in thousands of classrooms at once. And if you teach in Hindi or Marathi it's harder still: the ready-made question banks are written in English, for a different syllabus, and a general AI tool has never seen your textbook.

Approach

I built ShikshaOS around the book the teacher already has. Pick a textbook, pick a chapter, and it writes the quiz, or the full exam paper — sections A, B and C, up to 100 marks, answer key included — or the summary, or the lesson plan. It reads the chapter, works out which script it's written in, and stays in that language start to finish, so nobody has to pick a language from a dropdown. Each teacher gets a private workspace, and the whole thing is built to run at a cost a school can predict. It's running today, and I'm looking for schools to pilot it.

Impact

  • An evening of prep becomes a few seconds of generating and a read-through
  • Exam papers come out complete — sections, marks and answer key — not as a draft to assemble
  • Works in the language of the book, so teaching in Hindi or Marathi isn't a second-class experience
Showcase

Pipeline Operations Console

Oil & Gas · LLM

One console for a whole pipeline network — live monitoring, a failure-risk score on every segment, and month-end compliance in one click.

Showcase

Pipeline Operations Console

Oil & Gas · LLM

One console for a whole pipeline network — live monitoring, a failure-risk score on every segment, and month-end compliance in one click.

Problem

A pipeline rarely announces a failure. It thins and drifts for months first, and every one of those months leaves a trace in data somebody already owns. But the traces sit in six systems with six logins, and correlating them is nobody's job. So teams inspect on a fixed calendar, react when an alarm trips, and lose every month-end to paperwork.

Approach

I built one console that holds the whole network — forty segments or four thousand — on a single screen. The map is coloured by risk and refreshes every few seconds. Every segment carries a failure-risk score from seven factors, judged against its own history, so a reading that's ordinary on an old line still flags on a new one. Ask it a plain question and the answer comes back with the segments it read. Compliance evidence attaches to the segment as it's generated, so the monthly report is a filter, not an assembly job.

Impact

  • Risk scores move before the pipe does — no waiting on the inspection calendar
  • Alerts ranked by what they'll cost, not by what tripped first
  • Monthly compliance reports assemble from records already held
  • Ask the network a question, see which segments answered
Showcase

Crude Trade Copilot

Oil & Gas · LLM

A copilot for crude trading — read an assay (its specifications), work out the blend, predict the yield, all from a plain-language prompt.

Crude Trade Copilot

Oil & Gas · LLM

A copilot for crude trading — read an assay (its specifications), work out the blend, predict the yield, all from a plain-language prompt.

Problem

Buying crude oil starts with its assay — a detailed breakdown of what the crude contains. From there a trader has to work out how to blend it and predict what it will yield once refined. It's specialist work, done under time pressure, and a wrong call moves real money.

Approach

I built a copilot that reads the assay, works out the blend, predicts the yield, and writes a clear summary ready for a decision — all from one plain-language request. Work that used to take a specialist hours now happens in a single step.

Impact

  • Days of work assessing the right crude, now done in minutes
  • Consistent results every run — no prompt-wrangling, no guesswork
  • Fast enough to move on a high-margin spot crude before the window closes

Agentic Reporting Platform

Agentic AI · LLM

An agentic platform that turns weeks of client reporting into hours — AI agents read a team's own data and draft exec-ready output, leaving a person to review, not assemble.

Agentic Reporting Platform

Agentic AI · LLM

An agentic platform that turns weeks of client reporting into hours — AI agents read a team's own data and draft exec-ready output, leaving a person to review, not assemble.

Problem

Client reporting is slow, repetitive work. Pulling the numbers, assembling the narrative, formatting it for an executive audience — it can eat weeks, and the people doing it are the same specialists you'd rather have solving the underlying problem.

Approach

I led an agentic platform where AI agents read across a team's own data, work out what the report needs to say, and draft it exec-ready, end to end. Work that used to take weeks of assembly now takes hours, with a person in the loop to review and sign off rather than build from scratch.

Impact

  • Client reports in hours that used to take weeks
  • Specialists review and sign off instead of assembling from scratch
  • Exec-ready output drafted from a team's own data, end to end

Field Mobility Apps

Field Operations · Mobile

Mobile apps that carry multimodal field data — photos, notes, audio, video — back to a central server, then return the analysis as clear action points.

Field Mobility Apps

Field Operations · Mobile

Mobile apps that carry multimodal field data — photos, notes, audio, video — back to a central server, then return the analysis as clear action points.

Problem

Field teams see problems first, and they capture them however they can — a photo, a scribbled note, a voice memo, a quick video. But that's usually where it stops: the richest signal about what's happening on the ground sits trapped on a dozen phones, never reaching the people who could act on it.

Approach

I built mobile apps that let field professionals capture multimodal data — images, text, audio, video — and send it straight to a central server for analysis. From there it flows two ways: management gets the picture they need, and the field team gets specific action points back. The loop from observation to decision to action finally closes.

Impact

  • Field observations reach a central analysis instead of staying stuck on phones
  • Management sees what's happening on the ground without waiting for a report
  • Field teams get concrete action points back, not just an acknowledgement

Smart Education Apps

Education · Mobile + AR

Mobile apps that connect printed textbooks to digital learning — point a phone at a page to unlock videos, quizzes, and augmented content. One of the first of its kind in the country.

Smart Education Apps

Education · Mobile + AR

Mobile apps that connect printed textbooks to digital learning — point a phone at a page to unlock videos, quizzes, and augmented content. One of the first of its kind in the country.

Problem

A printed textbook is fixed the day it goes to press. Everything that makes a subject click — a video, a quick quiz that checks you actually understood, a model you can turn around in 3D — lives in a different world entirely, on a screen the book has no way to reach.

Approach

I built mobile apps that turn a printed page into a doorway. Point a phone at the book and it brings up the video, the quiz, the augmented content tied to exactly that lesson. It was one of the first projects of its kind in the country — printed and digital learning finally pointing at the same thing.

Impact

  • A printed page becomes a launch point for video, quizzes, and augmented content
  • Students keep the book they already have — the digital layer meets them there
  • One of the first printed-to-digital learning projects in the country

Want to hear more?

These are a few of the things I've built, with more on the way. If you've got a messy problem of your own — or just want the longer story behind one of these — I'd like to hear from you.

Get in touch
05Approach

How I take AI from idea to something people rely on.

01
Discovery
Find the real problem worth solving — and whether AI is even the right tool for it.
02
Architecture
Design how the whole solution holds together, and how it feels to use day to day.
03
Model selection
Pick what actually fits the job — local, on-prem, or cloud — not what's fashionable.
04
Integration
Wire it into the systems and data people already work in, so it lands in real use.
05
Evaluation
Build the frameworks that track answer quality and time-to-decision as people use it.
06
Rollout
See it through to daily, trusted use — and keep refining once it's live.
06About

What separates an AI demo from a system people trust.

01

Beyond the demo

Finding the real problem, proving the thing actually works, and getting it into people's hands. That's where the work is.

Beyond the demo

Most AI projects stall at the demo. I care about the part that comes after — the unglamorous work of turning a promising model into a system people open every morning and trust. So I architect the whole path: finding the real problem, designing how the solution holds together and how it feels to use, choosing the model, wiring it into the systems people already work in, building the evaluation that proves it works, and seeing it through rollout.

02

What I'm building now

Consoles and copilots that run where the data lives — sometimes on the device itself, so nothing sensitive leaves the building.

What I'm building now

Right now that's a set of AI solutions for live industrial operations: a console that lets a team reason over scattered inspection data instead of hunting through every file, a copilot that turns a specialist's workup into a single plain-language prompt, and a dual-path setup that runs inference on the device itself, so sensitive data never has to leave the building. The industry behind them matters less than the hard part they share — turning volume and contradiction into a call someone can stand behind.

03

Fifteen years of it

A decade of it went into digital twins — working simulations of real pipelines, used to keep them safe and running.

Fifteen years of it

The years before that went into software for other people doing consequential work, across industries that look nothing alike. A decade of it was digital twins — working simulations of real pipelines, used to keep them safe and running. The rest ranged wide: big-data engines reading over a million data points a second and analytics across billions of records for defense, mobility apps built for field professionals, education apps that bring printed books to life on a screen, finance apps, and IoT backends for connected devices. Different domains, different users — but the thread is the same, and it's the part I genuinely enjoy: the data shows up messy and contradictory, and what matters is whether someone can look at what the system says and make a confident call. Architecting the path to that moment is the work I find most satisfying.

07Contact

Have a problem worth solving? Let's talk.

LinkedIn

in/varunnidhi

↗

GitHub

@van007

↗

Based in India