I'm a CS graduate from Kolkata, India, and an AI safety researcher working on LLM cooperation and game-theoretic multi-agent systems. I've had work published at ICLR and NeurIPS, and I'm currently on a research fellowship studying how language models coordinate under conditions of strategic interaction.

When I'm not doing research, I build things — mostly for myself, occasionally for a few hundred thousand people.

A static, fully client-side route finder for Kolkata's buses. Route information here lives in people's heads and fragmented lists — this turns that into a searchable graph with direct, one-change, and two-change routes, autocomplete, and stop maps. No backend. No login. Just works.

Went mildly viral. 300k+ views.

HTML Python Graph search

A government scheme finder for Indian citizens, in 12 languages. Most people who need welfare schemes can't find them — buried in bureaucratic portals that assume digital literacy and Hindi fluency. This tries to fix that.

Multilingual NLP Civic tech

I have a habit of building small, specific tools for myself rather than looking for an app. Some of these:

• Insta Transcriber — pulls all my Instagram videos and transcribes them locally. Built it for my own content workflow.
• Flickr GIF — generates Anki-encoded flickering GIFs so I can learn Q&A pairs in a format that actually sticks for my brain.
• Learn French — a small app I built to study French, which I passed with a strong result after a compressed prep session.
• Book2Audio — converts books to audio. Built for personal use.
• Email Generator — context-aware draft generation. Also personal.
• FIFA 26 WC Prediction — a World Cup outcome predictor I'm actively working on, built around a Dixon-Coles base model with feature adjustment layers.

Most of these aren't public. I build for the problem, not the portfolio.

I work at the intersection of AI safety, cooperative AI, and game theory. My current focus is on how LLMs exhibit similarity-based cooperation — what it means, how it emerges, and what it implies for alignment.

• ICLR Oral (equal first author)
• NeurIPS publications
• h-index 5 · 233 citations (Updated June, 2026)
• Work cited in the 2026 International AI Safety Report

Languages — Python, JavaScript, HTML/CSS
ML/AI — PyTorch, Transformers, scikit-learn, NumPy, Pandas
Infra — Vercel, Docker, Git
Other — Flask, MySQL, Plotly

I'm figuring out how to be a researcher and a human at the same time. It's going okay.