Tell

A poker coach that questions your reasoning before revealing the math, then remembers your leaks across sessions via vector search, catching the same mistake even in differently worded hands.

Comment

Inspiration

I enjoy playing poker and I have always been interested in game theory and decision making. I wanted to get better at poker myself, so I decided to build something that could help me learn from the hands I actually play.

Instead of just telling me whether a decision was right or wrong, I wanted something that would make me think about my decision first. That is where the idea for Tell came from.

What it does

Tell is an AI poker study partner that helps you learn from your own hands.

You give Tell a hand you played and it walks through the important decisions with you. It asks what you were thinking, lets you explain your reasoning, and then shows you the math and explains what you could have done differently.

It also remembers the mistakes you make so it can identify patterns and help you work on the areas where you need the most improvement.

How we built it

We built Tell with React and Vite for the frontend and Python with FastAPI for the backend.

The backend runs on Google Cloud Run and we use Gemini for the AI coaching. We also built a separate poker math engine that handles things like pot odds and equity. This keeps the important calculations deterministic instead of relying on the AI to do the math.

We use Firestore to store hands, progress, and recurring mistakes. We also added voice input and text to speech so users can interact with Tell without everything having to be typed.

Challenges we ran into

The biggest challenge was learning the Google Agent Development Kit and figuring out how to properly integrate it into the project.

A lot of time also went into debugging. Getting the frontend, backend, AI, database, authentication, and voice features all working together took longer than expected.

Accomplishments that we're proud of

We are really happy with how the UI turned out. We wanted Tell to feel like an actual product and not just another AI chatbot.

We are also proud of the AI coach. It is genuinely useful for thinking through poker decisions instead of simply giving you an answer.

Getting voice, text, and the interactive poker UI working together was a lot of work, but we think it was worth it because it makes Tell much more enjoyable to use.

What we learned

We learned a lot about building AI agents and turning an AI model into part of a real product.

One of the biggest lessons was that AI does not have to do everything. Using code for things that need to be accurate, like poker calculations, and using Gemini for things like conversation and coaching made the system much more reliable.

What's next for Tell

The next step is getting Tell in front of real users.

We want to have more poker players test it, see how they actually use it, collect feedback, and improve the coaching experience based on that feedback.

Built With

Share this project:

Updates

Submission history