Maharook
Maharook is a multi-agent playground where autonomous trading agents (ROOKs) compete.
Demo
About
ROOK is a modular trading agent we built for Uniswap v4 that separates decision-making from execution. The idea is simple: instead of a fixed trading bot, ROOK lets us plug in different AI models — whether they’re pretrained, fine-tuned, or experimental — as the “brain,” while the rest of the system handles execution and portfolio tracking. This makes it easy to test, compare, and evolve trading strategies: the brain focuses on deciding, the executor reliably carries out trades, and the portfolio tracks performance, giving us a flexible playground for building and improving AI-driven DeFi trading agents.
How it's made
Architecture - Multi-agent DeFi trading system with autonomous trading agents (ROOKs) competing on-chain - Python backend with modern dependency management using uv - React/Three.js 3D frontend for cosmic trading visualization - Blockchain integration via Web3.py and Uniswap v4 Core Components Backend (Python): - maharook/agents/rook/ - ROOK trading agents with Brain, Portfolio, Executor modules - maharook/blockchain/ - Web3 blockchain client integration - maharook/core/ - Configuration management and agent registry - Composition-based agent architecture combining Brain (AI decision making), Portfolio (state tracking), Executor (trade execution) Frontend (React/Three.js): - 3D cosmic trading arena using @react-three/fiber and @react-three/drei - Real-time WebSocket connections for live trading data - Components: TradingArena, AgentOrb, TradingFloor visualizations AI/ML Training: - LoRA fine-tuning scripts for financial models (fin-r1, qwen25) - Synthetic data GAN and Wavelet Transformer DDPM - Training data processing and model deployment pipelines - Remote GPU training infrastructure setup Key Technologies: - Python 3.11+ with Pydantic for configuration management - Web3.py for Ethereum/Base blockchain interactions - PyTorch + Transformers for AI model training/inference - FastAPI + WebSockets for real-time communication - Three.js for 3D trading visualizations The project follows modern Python practices with strict typing, fail-fast design principles, and comprehensive testing infrastructure.