AI Investor – Multi-Agent Trading System
Developed a multi-agent trading system as part of MSc Computer Science thesis, integrating LLM-driven agents specialized in market sentiment, technical indicators, and risk management.
Project Overview
Developed a multi-agent trading system as part of MSc Computer Science thesis, integrating LLM-driven agents specialized in market sentiment, technical indicators, and risk management.
Each agent collaborated to generate and validate trading signals, forming consensus before execution.
Tested on simulated and historical stock market data to evaluate profitability, risk, and decision efficiency.
Challenges & Solutions
Challenges
- Coordinating reasoning across multiple autonomous agents
- Managing inconsistent LLM outputs in volatile data conditions
- Designing effective evaluation metrics for trading strategies
Solutions
- Implemented collaborative agent architecture using LangChain’s multi-agent framework
- Introduced consensus-based decision logic for reliable trade signals
- Used vectorbt for efficient backtesting and performance visualization
Results & Impact
Signal Quality
Agents produced more consistent buy/sell signals compared to single-model baselines
Risk Reduction
Reduced simulated portfolio drawdown through consensus filtering
Research Outcome
Validated potential of multi-agent LLM frameworks in algorithmic trading environments