Poly Predictor Kit

Logo

This project is basically a full Polymarket analysis toolkit we built for ourself because nothing out there does what we actually need. Everything is running locally, fast, and fully automated — no bullshit UI or slow manual work. You open any Polymarket event, hit a button, and you instantly get real insights.

The toolkit has three main features:

AI Market Insight (Gemini-based) This feature grabs all event + market data from Polymarket’s Gamma API and feeds it into Gemini to generate a short, simple breakdown of what the market is signaling right now. Not financial advice — just a clean summary of expectations, risks, and scenarios.
Emotional vs Rational Comments Classifier We trained a custom ML model on real Polymarket comments with SnowFlake integration. It pulls all comments from the event and shows whether the market sentiment is more emotional or rational. The idea is simple: emotional markets pump/dump harder, rational markets behave differently. It gives us a quick read on how noisy the event is.
Steamroller Detector This checks if a market has a “steamroller” structure: tiny upside vs massive downside. Basically tells you if one side is a trap. It uses liquidity, implied probabilities, wipeout factor, and time-to-resolution.

⸻

Tech stack • Chrome extension front-end • Local Python backends (Flask) • Gamma API for Polymarket data • Gemini API for insights • Custom ML model (sklearn) for sentiment • Comment scraper + automatic labeler • All running locally on 3 ports (5000, 5001, 5002) • SnowFlake ⸻

Why I built it

Polymarket is fun, but the data is messy. Manually checking comments, price shape, risks, expectations — it’s slow and inconsistent. So we built a system that does the boring part for me, and I only look at clean insights.

We're planning to keep improving it, add more models, more signals, better scraping, and eventually turn this into a full analysis kit.