Investment Committee
Five AI analysts with different investment styles research a stock with live data, debate it, and a chair weighs in.
Pick a U.S.-listed stock. Five AI agents, each with a different investment style and its own data tools, research it independently, then debate. A sixth agent chairs the committee: it scores each analyst's evidence, logic, and handling of risk, and gives the committee's view. Every number an analyst cites is checked against the data it pulled.
The agents
- The Alt-Data Scout: Looks for signals the market hasn't priced yet: prediction markets, insider buying and selling, and fresh SEC filings.
- The Fundamental Analyst: Starts from the business: revenue, earnings, valuation, and growth from the company's filings, checked against price trends.
- The Filings Reader: Reads what the company itself told regulators, in the latest 10-Q, 10-K, and earnings release, and quotes it word for word.
- The Sentiment Tracker: Reads the mood: what the news is saying, how the tone is shifting, and whether the narrative is turning.
- The Risk Quant: Thinks in probabilities and downside: volatility, value at risk, option-implied odds, and Monte Carlo scenarios.
- The Committee Chair: Has no data tools. Scores each analyst on evidence, logic, and risk, then gives the committee's view.
How this was built
I first built this for FISD's New York meeting in December 2025, as the first agentic AI demo the market data community had seen there: AI agents with different investment styles debating a stock live, each pulling its own data.
This version rebuilds it on Amazon Bedrock AgentCore with the Strands Agents SDK, and rethinks the data. Fundamentals, filings, and insider trades come straight from SEC EDGAR; prices from Alpha Vantage; news tone from GDELT; macro odds from Kalshi, a CFTC-regulated exchange, and company odds from Polymarket; the risk-free rate from the U.S. Treasury. The Risk Quant's options math uses QuantLib.
For a market data audience, provenance matters as much as the answer. Every source is shown with its timestamp; the code checks each number an analyst cites against the data the agents retrieved, and each quotation word for word against the filing it came from, with a link that opens the filing at that sentence.
"Agentic AI and Market Data" live demo, FISD New York, hosted by UBS, December 2025. Runs on DeepSeek V3.2 in Amazon Bedrock.
An educational demo of AI agents, not investment advice. The analysts are AI models that can be wrong; nothing here is a recommendation to buy or sell any security.
This demo is interactive and needs JavaScript.