Overview
Prediction markets like Polymarket let you trade on the outcomes of real-world events: elections, geopolitical conflicts, crypto price milestones, sports outcomes. Prices represent the crowd's probability estimate: a contract trading at $0.65 implies a 65% chance of "YES."
The opportunity? The crowd isn't always right. When breaking news drops at 3 AM UTC, prices lag. When a market requires domain expertise most traders lack, mispricings persist for hours.
My thesis was simple: could I build an AI system that reads faster, reasons deeper, and reacts quicker than the average Polymarket participant, while managing risk well enough to survive the inevitable wrong calls?
What I Built
A fully autonomous trading system with 24,307 lines of production Python across 36 modules, featuring:
- Dual-AI adversarial pipeline: Gemini 2.5 Pro as the analyst, Claude Sonnet 4.6 as the devil's advocate validator
- LMSR-aware execution: accounts for actual price impact instead of naive mid-price
- 6-layer risk management: circuit breakers, Kelly criterion sizing, drawdown controls
- Real-time dashboard: SSE-powered Flask app with 8 tabs and 13 API endpoints
- 487 automated tests across the full stack
- Production deployment on a VPS with systemd, nginx, and CI/CD
The Dual-AI Pipeline
Most AI trading bots use a single model. I deliberately chose an adversarial pipeline: one model acts as the analyst, the other as the devil's advocate. This mirrors how institutional trading desks work: the analyst pitches, the risk committee challenges.
Gemini 2.5 Pro was selected for its Google Search grounding: it can search the web in real-time during analysis. When evaluating a geopolitical market, Gemini searches for the latest dispatches, satellite imagery reports, and expert analysis at inference time.
Claude Sonnet 4.6 receives Gemini's analysis and challenges it. If Gemini says 80% confidence, Claude might counter with overlooked context and downgrade to 65%. This adversarial dynamic catches single-model blind spots.
Read the Full Technical Deep Dive
The full article covers LMSR market microstructure math, the complete system architecture with diagrams, all 6 risk management layers, the backtesting engine, security architecture, and lessons learned.