Multi-Agent Research: How 6 LLM Teams Analyze 900 Stocks

This article describes an autonomous trading system that uses LLM agents for research, ML for pattern recognition, and deterministic rules for execution. It explains how the system distributes 900 stocks across 6 sector-specialized teams to screen, analyze, and debate the best opportunities.

Details

The article introduces Nous Ergon, an autonomous trading system that splits intelligence across four layers: LLM agents for research judgment, ML for pattern recognition, deterministic rules for execution, and a backtester for system-wide learning. The focus is on the Research module, where LLM agents work in parallel to analyze a universe of ~900 mid-to-large-cap US stocks. Each of the 6 sector-specialized teams follows a 4-stage pipeline: quant analysis, qualitative review, peer collaboration, and CIO evaluation. The quant analyst uses screening tools to narrow the ~150 stocks in their sector down to a top 10, while the qual analyst adds deeper context from news, filings, and past signal failures. The teams then agree on 2-3 final recommendations, which are evaluated by a CIO agent across factors like conviction, macro alignment, and portfolio fit. This process results in a ~25-stock portfolio that is refreshed weekly. The article also provides an example of a LangChain tool used by the quant analysts to access balance sheet data.