01
Work
The last couple of years I've been building products at places working on problems I find interesting.
Founding Engineer
Dec 2024 — presentRaccoon AI
- •Shipped Elemetry, a cache engine that stores browser actions in an in-memory vector graph database, leveraging browsing history to accelerate repetitive tasks
- •Designed dual-model (ephemeral events vs. persisted blocks) real-time streaming pipeline for frontend rendering of realtime agentic work
- •Built a text-to-video motion-graphics system with a React rendering engine, director's-board approval flow, parallelized frame-by-frame export on serverless GPUs, and a full audio toolkit
- •Also shipped canvas editor, raccoon-inspector, worked heavily on chat app's state management and performance
Joined as an intern, now founding engineer. Raccoon builds browser agents that remember what you did yesterday and skip the repetitive clicks today. I worked on Elemetry: a cache engine that stores browser actions as a vector graph, so the agent can replay past browsing patterns instead of working them out from scratch every time. Most of my time went into getting the graph structure to capture intent rather than just actions.
Research Intern
Jul 2023 — Dec 2024IIT Bombay · EdTech Research
- •Built adaptive learning system with video bots for student engagement
- •Implemented feedback platform — React, Express, Flask running deep learning model
- •Shipped as Docker container and Electron app for offline use
- •Lead backend developer on GeoGuru — LMS with geographical information system
Started as a summer research intern and stayed on. The affect-aware system uses video bots that adapt based on how engaged students are — the deep learning model runs on Flask and feeds back into the React frontend in real time. GeoGuru took longer: a full LMS that integrates QGIS for spatial data. I work on the backend architecture, which has to handle concurrent map rendering sessions.