Walsh: Multi-agent research pipeline with risk manager that can veto trades

Walsh is a multi-agent AI stock research system that analyzes tickers through four specialized agents (Fundamental, Technical, Sentiment, and Macro) in parallel, synthesizes their recommendations via a confidence-weighted Portfolio Manager, and routes final decisions through a Rule-Based Risk Manager that can veto trades. In a 2023 backtest on AAPL, MSFT, and NVDA, the system returned +75.52% with a 2.36 Sharpe ratio, underperforming buy-and-hold (+119.75%) due to its conservative risk thresholds causing it to exit positions during periods of high uncertainty and miss subsequent rallies. The system is designed to prioritize downside protection over maximum returns, with pending evaluations on per-agent accuracy calibration and cost/latency metrics once LLM-based agents replace the current rule-based implementation.