ΣIGMA

Dashboard showing messages interacting and graphs appearing
Final Decision Dashboard
The Analysed Reports

ΣIGMA — Multi-Agent AI Trading Analyst 💡 Inspiration Hedge funds succeed because they operate as teams of experts — technical analysts, fundamental researchers, risk managers, and sentiment trackers all working together to make smarter decisions. We wanted to democratize that edge. What if anyone could have a 24/7 AI trading team — each agent specializing in a domain, analyzing data, and debating in real time? That’s the vision behind ΣIGMA (Sigma) — an AI system that combines diverse analytical minds into one cohesive intelligence.

⚙️ What It Does ΣIGMA is a multi-agent AI trading analysis platform where six specialized AI agents collaborate to produce a unified trading decision: Market Analyst — Studies RSI, MACD, Bollinger Bands, and SMAs for momentum and trend. News Analyst — Parses live headlines to understand sentiment and catalysts. Fundamentals Analyst — Examines P/E ratios, revenue growth, margins, and ROE. Payment Flow Analyst — A unique agent analyzing transaction volume, fraud rates, and ecosystem signals. Risk Manager — Synthesizes all agent insights to assess risk and reward. Trader — Makes the final decision (BUY / SELL / HOLD) with a confidence score. Type a ticker (like AAPL or TSLA), and watch the agents analyze, debate, visualize data in Plotly charts, and arrive at one clear call.

🛠️ How We Built It Backend (Python): FastAPI for async APIs and WebSockets LangGraph for orchestrating the six-agent workflow Anthropic Claude Sonnet 4.0 for analysis and Claude Haiku for summarization Data Sources: Yahoo Finance, StockStats, Google News RSS, and a custom payment flow simulator Frontend (React + TypeScript): Built with Vite, Tailwind CSS, and shadcn/ui for sleek styling Interactive charts powered by Plotly.js Real-time agent updates via WebSockets with an HTTP polling fallback for reliability Optimization: LangGraph originally generated over 7,000 messages per run — we filtered, summarized, and reduced this to six concise, high-signal updates per analysis.

🧱 Challenges We Ran Into Message Overload: Thousands of redundant logs → solved via filtering + summarization. Async Deadlocks: Fixed generator exhaustion bugs in LangGraph streams. WebSocket Failures: Implemented hybrid polling for guaranteed message delivery. API Costs: Added “demo mode” using cached AAPL data for zero-credit testing. Plotly Rendering: Large chart HTML handled safely with sandboxed iframes. Frontend Errors: Implemented fallback parsing for unpredictable backend responses.

🏆 Accomplishments That We're Proud Of Built a fully functional multi-agent orchestration system using LangGraph. Achieved real-time collaboration between autonomous LLM agents. Created a production-ready architecture with async streaming and resilience. Reduced verbose model outputs into human-friendly conversations. Developed a demo mode that showcases ΣIGMA instantly — no API credits required.

📚 What We Learned Multi-agent design works best when each agent has a defined, minimal role. Summarization and filtering are key to user-friendly AI systems. Reliability matters: WebSocket + polling ensures no message loss. Efficiency and clarity beat complexity — especially in AI systems. Demo environments are vital for testing and pitching effectively.

🚀 What’s Next for ΣIGMA Integrate live trading APIs (Alpaca, IBKR) for paper trading and execution. Add agent debate and consensus to simulate team-style discussions. Expand coverage to crypto, ETFs, and commodities. Launch a public dashboard where anyone can access institutional-grade analysis. Develop lightweight domain-tuned models for lower latency and cost. In short: We turned the complexity of hedge fund research into a real-time, explainable, AI-powered trading system — where six agents think, argue, and decide together. That’s ΣIGMA — the sum of intelligence.