> SYSTEM OVERVIEW
// RL AGENT (Neural Network)
Reinforcement learning agent trained on 50,000+ trades from 15-minute crypto prediction markets.
- Strategy: Fast in-and-out momentum trading
- Markets: BTC/ETH 15-minute up/down
- Avg hold time: 2-8 minutes
- PyTorch neural network with attention
- Rule-based exits (TP/SL/Time stops)
// OPUS AGENT (AI Research)
AI-powered agent using Claude + web search to find mispriced markets across ALL Polymarket.
- Strategy: Research-based edge finding
- Markets: Politics, sports, crypto, events
- Scans 3000+ markets every 30 minutes
- Real-time web search for news
- Claude AI probability estimation
> TRAINING DATA
// Training Stats
- Episodes trained: 50,000+
- Simulated profit: $50,000+
- Training method: PPO with experience replay
- Features: Orderbook, momentum, volatility
- Binance futures correlation signals
// Risk Management
- Max position: 15% of bankroll
- Stop loss: Dynamic based on entry
- Take profit: Trailing stop system
- Max hold time: 12 minutes
- Liquidity protection: Auto-exit
> TECHNICAL STACK
// Infrastructure
- Python 3.12 + asyncio runtime
- PyTorch (CUDA) neural networks
- Real-time WebSocket connections
- Cloud-native database sync
- Edge-deployed static frontend
// Data Sources
- Polymarket CLOB (orderbook + execution)
- Binance Futures (BTC/ETH/SOL correlation)
- Real-time web search engines
- Claude Opus (AI research + analysis)
- Custom sentiment aggregation
This is an experimental trading system running with real money on Polymarket.
Past performance does not guarantee future results. The agents can and will lose money.
This is not financial advice. Use at your own risk.