AlphaQuant
Polymarket analytics dashboard that combines prices, language model output and statistical models into one probability estimate.
Overview
A personal project: a Polymarket analytics dashboard that combines market prices, language model output and statistical models into one probability estimate. It flags markets where that estimate differs from the market price.
Data pipeline
A scheduler runs every 10 minutes, fetches live markets from the Polymarket Gamma API and stores them in SQLite. Each market is sorted automatically into a category such as macro, crypto, tech, politics or sports.
Probability engine
Llama-3 on Groq reads recent news and returns a probability and a sentiment score. A statistical baseline adds signals such as mean reversion, volume, price uncertainty, momentum and a category prior. The final probability is a weighted mix of the market price, the language model and the statistics, with the language model weight scaled by its confidence; expected value is the final probability minus the market price.
Dashboard
A FastAPI backend serves markets filtered by category and minimum expected value, plus charting endpoints. I built the historical data pipeline, the charting endpoints and the dashboard interface with timeframe controls and market search.
Next
Still in progress: probability calibration, backtests scored with Brier score and log loss, and a machine learning model trained on market history.