I build systems at the seams, where the backend, the model and the interface have to agree.

Computer science at VIT Vellore, based between Vellore and Abu Dhabi. Most of what I work on sits in the gap between a model that works in a notebook and a system that survives being run twice at once.

Currently: a governed AI platform, a Shopify operations copilot built on LangGraph, an e-commerce build for a client in Nepal, plus a real-estate intelligence API from a recent Hub71 hackathon.

82%cost reduction in the HIRE4GIG guarantee engine

38msper matching window across 50,000 orders

0.949ROC-AUC, three-model pneumonia ensemble

1patent application filed

Selected work

HIRE4GIGpatent filed

Income guarantees for gig workers that adapt every three minutes, matched against orders with a fairness-aware cost function.

Python · Gaussian process regression · Hungarian algorithm

A Shopify operations copilot where the model plans and deterministic Python executes. Every write is gated behind a human approval it cannot route around.

LangGraph · FastAPI · MCP · AGPL-3.0

Electronics storefront, Nepalclient

Full-stack commerce platform for the Nepali market: NPR pricing, connectIPS payments in progress, typo-tolerant Postgres search, and an admin panel scoped to the website rather than the business.

Next.js 16 · TypeScript · Prisma · Supabase

render-langgraph

Renders a LangGraph topology in the browser with no code changes and no LangSmith account. Published on PyPI as a v0.1 walking skeleton.

Pneumonia detection ensemble

Three CNNs averaged over 5,863 chest X-rays, with Grad-CAM heatmaps so a radiologist can see what the model looked at.

TensorFlow · DenseNet121 · ResNet50 · MobileNetV2

Recent writing

Two copies of my agent woke up owning the same task

The database that accidentally kept my agent alive last time became a liability the moment I ran two instances of it. Leases, fencing tokens, and the one integer that stops two writers from lying to each other.

I almost fixed the exact thing keeping my agent alive

Two days into fixing a bug, I realized the thing I was about to remove was the reason my agent worked at all. The in-memory checkpointer I trusted was doing nothing in production. A database write I'd added for a different reason was quietly reassembling the whole pipeline from scratch.