EquityTraderX: Two-Layer Multi-Agent Trading System

• May, 2026

We built a two-layer multi-agent trading architecture combining CrewAI and LangGraph. Layer 1 utilizes a 5-agent CrewAI crew to programmatically generate the trading platform's codebase from a natural language specification. Layer 2 executes an 11-node LangGraph pipeline featuring parallel analyst fan-out (News, Fundamentals, Technical) and a dynamic coordinator with hybrid Python-LLM risk enforcement. We evaluated the system using Claude, Gemini, and Grok models across a five-day live paper trading simulation. The top-performing configuration achieved a 21.17% total return and a 19.418 Sharpe ratio. We also implemented a Reflection Agent to assign groundedness scores and deployed a Streamlit dashboard with an SQLite audit trail for real-time performance tracking.