I build agents that do real work outside the chat window: on electronics lab benches, in small-shop ledgers, and across the SaaS tools teams already use. Based in Chennai.
Open source
- typescript-eslint: fixed the AST type for JSX fragments used as attribute values (#13051, shipped in v8.72.0)
- mocha:
--fail-zeroreturned the wrong exit code in parallel mode (#6433, in review)
Things I've built
- ventriloquist: turns Mac apps into MCP servers through the accessibility API. After the first run there are no screenshots and no LLM in the loop.
- Causeway: on its lab corpus, turns 938 security alerts into 17 incidents, each with the root cause and the smallest action that would have stopped the intrusion. No model in the decision path, about 25 ms per run.
- ohmwork: answers open-ended electronics-lab questions with a circuit file and an analysis computed by a real simulator, not guessed by a model. Live
- keel: a churn-recovery agent that investigates risk across Stripe, Linear and usage data, then runs policy-gated, read-back-verified actions in Linear, Notion, Resend and Slack. Live
- KairoRM: point it at any repo and get architecture docs, module guides and a live Q&A terminal UI.
- hacksprint: a voice-first cash planner for kirana shop owners, with a weekly forecast, collections and supplier-bill reading. Live
Now
Contributing to open-source TypeScript tooling and AI-agent infrastructure. Open to internships at early-stage startups building agents or developer tools.