I build AI agents and tools that solve problems I run into — then occasionally they solve problems for other people too.
By day I work on VMware Cloud Foundation at Broadcom. On my own time, I've been deep in AI agents and frameworks — building things like an autonomous bot that runs growth for a streaming search engine, an MCP server that lets LLMs talk to any API, and a pipeline that digitized a 149-page historical manuscript about my family's village in Bihar.
I ship things. Some have thousands of users. Some have an audience of one. The best ones started as a personal itch.
MCP server that lets LLMs discover and call any API dynamically.
Instead of loading entire API specs into context, JitAPI uses semantic search and dependency graphs to find only the endpoints needed for each task. Register any OpenAPI spec, and LLMs can query and execute multi-step API workflows automatically. Published on PyPI, works with Claude Code and Claude Desktop.
AI search that finds what to watch and where to stream it.
I built this because I was wasting 30 minutes every night deciding what to watch. WhatToStream is an AI search that tells you where to stream any movie or show across Netflix, Prime, Hulu, and more. Search like "Horror movies under 2 hours on Netflix" — it understands intent, not just keywords. Unlike Google, it queries verified streaming data directly. No SEO rabbit holes, no affiliate clickbait.
Autonomous AI agent that runs growth for WhatToStream — 24/7, no human in the loop.
After building WhatToStream, I needed someone to run it. So I built PunkClanker: an autonomous AI agent running on a Mac Mini. It handles SEO, ships features, fixes bugs, generates pages for trending content, and posts to social. No standups. No handholding. It grew search visibility 36x and generated over 6,800 pages autonomously.
Chat with your Apple Health data in plain English.
Export your Apple Health data, then ask questions like "How many runs this year?" or "What was my average pace in November?" Under the hood it uses RAG over SQL — indexing your health data and using an LLM to translate natural language into queries. Started because I wanted to make an Instagram post showing my running stats and realized Apple Health makes it nearly impossible to get simple answers.
AI agent for Home Assistant with ephemeral automations.
Home Assistant automations are persistent YAML files — but most of what you actually want is one-shot. "Turn on the porch light when I get home" just this once. This agent controls your entire HA setup via Telegram with natural language, and creates ephemeral tasks that fire once and clean up automatically. The HA community has been requesting this since 2020.
Digitizing a 149-page manuscript about my ancestral village.
Original manuscript
CHAPTER IV RELIGIOUS LIFE Chatha - divali - Sukarati (pakheva) - yamadvitiya - Godhan - Makara Sakranti - Sripancami - Sivaratri - Dola (Holi) - Ramanavami CHAPTER V ADMINISTRATION AND POLITICS Village pancayata - village headman (Chaudhuri) - Primary judicial system CHAPTER VI CONCLUSION APPENDICES 1. Folk songs of Morsand 2. Special terms 3. Photographic plates 4. Archaeological evidences
OCR + Claude transcription
My dad showed me a stack of yellowed manuscripts — 20 years of research by Prof. Harishchandra Satyarthi documenting our village Morsand in Bihar from the 17th century to 1947. The 149 pages were a mix of typewritten English and handwritten Hindi that had resisted manual transcription for years. I built a pipeline: high-res scans → Google Cloud Vision OCR → Claude cross-referencing against originals to correct errors and translate Hindi passages. Each page took 60-80 seconds. The manuscript maps migration genealogies, feudal labor systems, and evidence of a community on the site as far back as 500 BCE.