AI coaching from poker hand histories
A system of several AI agents that turns raw hand data into short, specific feedback that a player can act on.
- Client
- Online coaching platform
- Role
- Tech lead
- Methods
- Multi-agent orchestration, evaluation
The problem
A poker player produces a large number of hand histories. Solver software can already say what the correct play was in each hand. Neither tells the player what to work on.
The gap is between a correct answer and useful coaching. Which mistakes repeat? Which ones cost the most? And how do you say it so that the player changes how they play?
What I do
I am a tech lead on the product. I contribute to the agent architecture: the multi-agent system that reads the hands of a player and writes the feedback.
I have shipped several of the product's AI coaching features.
What this page leaves out
The design of the system and its results belong to the client, so they are not on this page.
For work that I can describe in full, see the reinforcement-learning case study and my own research on health records.