How the idea came?

TeetoSatya came from something that happens in everyday conversations. Someone says something confidently, another person says, "There is no way that's true," and suddenly everyone is searching on their phone. Sometimes nobody checks it and the conversation continues with information that might not even be true.

We started thinking about why checking a fact has to interrupt the conversation. What if it could happen while people were still talking? We turned that idea into a competitive game where people can challenge each other's claims and there is actually something to lose if you are wrong.

How we built it

We use LiveKit for the live conversation and ElevenLabs to transcribe the current speaker in real time. We keep the transcript connected to the person who actually said it. Gemini then finds factual claims from the conversation. When someone presses Fact Check, they can choose a recent claim to challenge. Gemini and Claude evaluate it. If the claim is incorrect, the challenging team gets 100 points and the speaker loses their turn. If the claim is correct, the challenger loses 50 points. We use SpacetimeDB to keep players, scores, turns, claims, speaker queues, and other game states synchronized for everyone.

Challenges we faced

One of the biggest challenges was figuring out who said what. Having a transcript was not enough. Every claim had to stay connected to the correct speaker. We also had to make several things happen together when someone pressed Fact Check. The timers had to stop, everyone had to see the same result, points had to update only once, and the correct speaker had to rotate.

What we learned

We learned that building something with real time voice and AI is not only about getting a response from a model. A lot of the work is making sure everything happens at the right time and every user sees the same thing. The debate game is our first way of showing TeetoSatya, but the idea is bigger than debate. We can imagine it being useful in classroom discussions, podcasts, interviews, group conversations, or even friends arguing about random things.

We want people to be able to check what they hear without having to leave the conversation to do it.

Built With

Share this project:

Updates

Submission history