Stuff I've been tinkering with on the side — mostly voice agents and small tools, usually built fast at hackathons or on weekends. Some live, some held together with tape. Growing list.
A VC brain. The framework you judge deals against, built as an agent — it researches a company live, scores it against a fund's thesis across separate axes, then argues both sides: the bull case, the bear case, and the one thing it can't figure out. Ends with the five questions worth asking the founder, and an honest list of what it isn't sure about. Being built now.
A voice-first trip planner. Say "5 days in Bali for two, 1.5 lakhs" and it builds a full itinerary — real flights, hotels, day-by-day activities, weather, live budget — in one conversation. Built on Cloudflare Workers + ElevenLabs Conversational AI, with Llama 3.1 for intent parsing, semantic search for fuzzy asks like "somewhere like Kyoto but warmer," and live price scraping.
A voice-first shopping agent with a personality. Tell her what you want, your budget and currency; she asks the right questions, searches the web live, and reads your top three picks back — say "buy it" and the product page opens. Co-built with Aishwarya Sudhakar on ElevenLabs Conversational AI (with VAD, interruptions, client tools), Firecrawl search, and Claude for intent parsing and ranking.
Built for a VC I wanted to talk to, as the pitch itself. Enter a company or founder name and it researches them live, maps them against that fund's thesis, checks portfolio adjacency, and outputs a structured prep brief. Bring your own key, nothing stored server-side. This was the first pass at the idea that became Rubric.