An experimental platform where AI-generated trading algorithms compete against each other in a simulated stock market. Multiple Large Language Models (LLMs) automatically generate trading strategies, which then battle in a 5-minute trading session to see which algorithm achieves the highest return on investment (ROI).
All agents start with the same capital and compete in real time as prices shift based on their combined trading behavior.
This platform demonstrates the intersection of AI code generation and algorithmic trading by:
- Using LLMs to generate trading algorithms - Select from 50+ AI models (Claude, Gemini, GPT, Llama, etc.) to automatically generate Python trading functions
- Simulating realistic market conditions - A custom order book and tick-based engine processes trades with price-time priority matching
- Competing strategies head-to-head - 2-6 AI-generated algorithms compete simultaneously on the same stock data
- Visualizing performance in real-time - Interactive dashboard shows live leaderboards, portfolio values, and market charts
Each generated algorithm receives tick-by-tick market data and makes autonomous BUY/SELL/HOLD decisions. The simulation engine processes these orders through a realistic order-matching system, and the agent with the highest ROI at session end wins.
- 50+ LLM Integration - Generate strategies using Claude Opus, Gemini, GPT-4, Llama, DeepSeek, Mistral, and more via OpenRouter API
- Automatic Code Creation - LLMs write complete Python trading functions based on strategy prompts
- Algorithm Preview - Review and inspect generated code before running simulations
- Multi-Model Comparison - Test which AI model creates the most profitable strategies
- Order Book Engine - Professional-grade order matching with price-time priority
- Realistic Tick Data - Historical data for AAPL, GOOGL, TSLA, MSFT, AMZN, NVDA, META, NFLX
- 60-Tick Sessions - Each battle runs for 60 market ticks (~5 minutes real-time)
- Multiple Agent Support - Run 2-6 competing algorithms simultaneously
- Live Leaderboard - Real-time ROI rankings updated every tick
- Market Charts - Dynamic price visualization with Recharts
- Performance Metrics - Track P&L, win rate, max drawdown, and portfolio value
- Trade History - Complete audit log of all executed orders
Each trading session lasts 60 ticks (approximately 5 minutes). Agents are ranked by:
| Metric | Description |
|---|---|
| ROI (%) | Return on investment - primary ranking metric |
| P&L | Absolute profit or loss in dollars |
| Win Rate (%) | Percentage of profitable trades |
| Max Drawdown (%) | Largest drop from peak portfolio value |
| Portfolio Value | Total cash + stock holdings at market price |
| Trade Count | Total number of executed trades |
The agent with the highest ROI wins the session.
| Component | Technology | Purpose |
|---|---|---|
| Web Framework | Flask | REST API server |
| Language | Python 3.11+ | Core simulation logic |
| LLM Integration | OpenRouter API | Access to 50+ AI models |
| Market Data | Yahoo Finance (yfinance) | Historical stock data |
| Data Processing | Pandas, NumPy | Data analysis |
| Server | Gunicorn | Production WSGI server |
| Component | Technology | Purpose |
|---|---|---|
| Framework | React 18.2 | User interface |
| Routing | React Router v7 | Page navigation |
| Styling | Tailwind CSS | Responsive design |
| Charts | Recharts 3.3 | Market visualization |
| Animation | Framer Motion 12 | Smooth transitions |
- Frontend Hosting - Vercel
- Backend Server - Flask with Gunicorn
- No Database - In-memory simulation state
The platform is live at: ai-trader-battlefield-fro.onrender.com
User selects AI models → OpenRouter API generates trading functions → Code saved as Python files
Each generated algorithm implements:
def execute_trade(ticker: str, cash_balance: float, shares_held: int) -> str:
# AI-generated trading logic
return "BUY" | "SELL" | "HOLD"Load tick data → For each tick:
- Call each agent's execute_trade()
- Collect orders (BUY/SELL/HOLD)
- Match orders in order book
- Update prices based on trades
- Calculate portfolio values
→ Return final rankings
- Price-time priority - Best price gets filled first; ties broken by timestamp
- Partial fills supported - Large orders can be partially executed
- Bid-ask spread - Realistic market microstructure
- No short selling - Long-only strategies
| Endpoint | Method | Description |
|---|---|---|
/api/ai_agents |
GET | List available AI models |
/api/data_files |
GET | List available stock datasets |
/api/algos |
GET | List generated algorithms |
/api/algos/{filename} |
GET | Preview algorithm code |
/api/generate |
POST | Generate new algorithms |
/api/run |
POST | Start simulation |
/api/status/{sim_id} |
GET | Get simulation status |
algoclash-v1-test/
├── backend/
│ ├── app.py # Flask server & REST API
│ ├── market/ # Simulation engine
│ │ ├── market_simulation.py # Session orchestrator
│ │ ├── order_book.py # Order matching engine
│ │ ├── tick_generator.py # Stock data provider
│ │ └── agent.py # Trading agent manager
│ ├── open_router/ # LLM integration
│ │ ├── algo_gen.py # Algorithm generator
│ │ ├── model_fecthing.py # Fetch AI models
│ │ └── ai_agents.json # 50+ model configs
│ ├── data/ # Historical stock CSVs
│ └── generate_algo/ # Generated Python algorithms
├── frontend/
│ ├── src/
│ │ ├── components/ # 16 React components
│ │ │ ├── Dashboard.js # Main battle arena
│ │ │ ├── ResultsDashboard.js # Leaderboard
│ │ │ ├── Models.js # AI model directory
│ │ │ └── ...
│ │ └── App.js
│ └── package.json
└── README.md
- Educational platform for algorithmic trading concepts
- AI code generation benchmark (which LLM writes better trading code?)
- Strategy testing environment with realistic market simulation
- Interactive demonstration of market dynamics
- Production trading system (no real money or live markets)
- Financial advice or investment tool
- High-frequency trading platform
- Connection to real brokerages
SimulationConfig(
max_ticks=60, # Session length (60 ticks)
tick_sleep=1.0, # Seconds between ticks
initial_cash=10000.0, # Starting capital per agent
enable_order_book=True, # Realistic order matching
allow_short=False, # No short selling
allow_negative_cash=False # No margin trading
)- AAPL (Apple)
- GOOGL (Google)
- TSLA (Tesla)
- MSFT (Microsoft)
- AMZN (Amazon)
- NVDA (NVIDIA)
- META (Meta)
- NFLX (Netflix)
Contributions are welcome! This is an experimental platform under active development.
Current Branch: v1
Main Branch: main
See LICENSE file for details.
- OpenRouter - For providing unified access to 50+ LLM providers
- Yahoo Finance - For historical market data
- All open-source libraries that made this project possible
Built with Flask, React, and AI curiosity.