VVarun NidhiAI Product Leader
Helping people make better calls under uncertainty
Scroll

Varun Nidhi

Varun NidhiVarun Nidhi

I build AI products
that turn messy operations into decisions.

I build AI products for the hard end of real operations — anywhere the data is incomplete, the conditions are unforgiving, and a wrong call is expensive. The job is always the same: get someone to a faster, better-grounded decision they can actually trust.

See what I've builtGet in touch
What I build
AI products that make it into real, daily use
Where it runs
On the device and in the cloud — privacy first
The field
Industrial operations — where conditions are hardest
01About

Fifteen years in industrial tech taught me one thing about building AI.

A model that works in a demo and a model that works in a live operation are not the same thing — and that gap looks the same in any industry where the data is noisy and the decisions carry weight. I build for the side that holds up in the real world.

Most AI projects stall at the demo. I care about the part that comes after — the unglamorous work of turning a promising model into something people open every morning and trust. So I own the whole path: finding the real problem, designing how it feels to use, choosing the model, building the infrastructure, and seeing it through rollout.

Right now that's a handful of AI products 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 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.

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: mobility apps built for field professionals, education apps that bring printed books to life on a screen, finance apps, and the big-data analytics that sit under all of it. 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 software says and make a confident call. Getting them to that moment is the work I find most satisfying.

Since 2010
Fifteen years in industrial tech — from wireless to AI
End to end
I own the whole path — problem, design, model, infrastructure, rollout
Across industries
Pipeline safety, field mobility, education, finance, and big data
02Things I've Built

Products built to survive contact with the real world.

Open any one for the problem I set out to solve, how I approached it, and what changed once it shipped.

Pipeline Operations Console

Oil & Gas · LLM

An AI console that helps teams reason over pipeline inspection and operations data — and turn it into action in one click. The LLM does the work; no one writes a single prompt.

Pipeline Operations Console

Oil & Gas · LLM

An AI console that helps teams reason over pipeline inspection and operations data — and turn it into action in one click. The LLM does the work; no one writes a single prompt.

Problem

Pipeline inspection and operations data shows up in volume and in pieces — survey files, inspection logs, spreadsheets, SCADA, and more. Making sense of it is slow, manual work, and the answer is only ever as good as whoever had the patience to read every file.

Approach

I built a console that reads across inspection records, operations data, and GIS data and makes sense of it together — so a team can ask a plain-language question, get an answer grounded in their own records, and turn it into action in one click. Evaluation loops track how good the answers are and how fast people reach a decision as they use it.

Impact

  • One-click reports replace hours and days of manual assembly
  • Teams reason over inspection and operations data by asking, not hunting through files
  • Adoption and answer quality tracked with built-in evaluation loops

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

Process Digital Twins

Oil & Gas · Simulation

Working digital models of oil & gas pipelines — used to simulate behavior and check metering before it matters in the field.

Process Digital Twins

Oil & Gas · Simulation

Working digital models of oil & gas pipelines — used to simulate behavior and check metering before it matters in the field.

Problem

You can't safely test how a pipeline will behave — or trust what its metering reports — by experimenting on the live asset.

Approach

I spent a decade building process and metering digital twins: working simulations of real pipelines, used to model behavior and validate metering against expected physics before it matters in the field.

Impact

  • Pipeline behavior simulated without touching the live asset
  • Metering validated against expected physics ahead of the field
  • A decade of integrity and operations work built on shared models

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
03Open Source

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

Small, self-contained tools — each one runs AI locally or right in your browser. Free to clone, fork, or just take ideas from.

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
04Contact

Have a problem worth solving? Let's talk.

Tell me what you're working on — or just say hello. Every message reaches me directly, and I read all of them.

LinkedIn

in/varunnidhi

↗

GitHub

@van007

↗

Based in India