Labs · Adam Honoré

Feed Doctor

A live market data feed, a few things going wrong, and an AI that explains them.

Watch a simulated real-time market data feed, inject the failures that keep operations teams up at night, and see how rules and AI work together: rules catch problems instantly, AI explains the likely cause, the client impact, and what to do next.

How this was built

I've spent much of my career around market data, where a two-second gap in a feed can mean a call from a trading desk. Detecting problems was never the hard part; feeds are instrumented to the microsecond. The hard part is the next five minutes: what does this mean, who's affected, and what do we do?

So this demo splits the job. Plain code detects incidents the instant they happen, because detection has to be fast and deterministic. AI does what people are slow at under pressure: explaining the likely cause, the client impact, and the next steps, and writing the shift handover at the end. The feed is simulated; no real market data is used.

Exploring AI for real-time data operations, October 2026. Runs on DeepSeek V3.2 in Amazon Bedrock.

This demo is interactive and needs JavaScript.