Pulseportfolio started because I wanted to track my own investments properly and none of the existing tools did quite what I wanted, at the price I wanted. The frontend is Next.js, the real-time data layer runs on Convex, and the heavier backend work(market data fetching, fault-tolerant multi-provider aggregation, AI routing) lives in Python FastAPI microservices. The chatbot uses a RAG pipeline so queries are grounded in your actual portfolio data rather than just guessing. I set up a full CI/CD pipeline with GitHub Actions running tests before deploying to Railway.
It's still very much a work in progress. The market data side is fragile in ways I'm not fully happy with. The RAG pipeline works, but the quality of responses degrades depending on how well the user's portfolio data is structured. There's a lot I want to do on the analytics side too from proper charting, performance breakdowns and comparison tools. It's a project I genuinely use myself, so it'll keep evolving.