3 weeks ago
Wed Jan 14, 2026 8:11pm PST
Show HN: I made a search engine for prediction markets
Prediction markets are having a moment, but searching them is painful: - Markets are spread across 9+ platforms - Each platform has different UX and search capabilities - No way to compare similar markets across platforms - Finding new markets on topics you care about requires manual checking

I built UPMI to solve this. One search queries all platforms, AI scores relevance, and you get a unified table of results sorted by match quality.

The architecture is fairly straightforward: - Firecrawl maps platform URLs in parallel - Gemini Flash fetches + analyzes actual market pages - Streaming SSE for progressive UI updates - Chat interface for deeper analysis on selected markets

What I learned: - URL context in Gemini is powerful but rate-limited (throttling required) - Structured outputs need careful schema design or you get hallucinations - Real-time streaming UX is hard but worth it for perceived speed

Tech: Next.js 16, React 19, Gemini API, Firecrawl, Neon Postgres

Try it: [URL]

Looking for feedback on relevance scoring and whether this is useful to actual traders vs. just a fun project.

read article
comments:
add comment
loading comments...