AI Debate Arena

Two AI personas argue any decision out loud, and you cast the final vote.

Comment

Inspiration

Sometimes you have a problem deciding on what to eat or where to go, or perhaps want some entertainment from a specific topic debate. The AI Debate Arena allows two AI personas to voice out their opinion in a minimum of 3 chances each, and you judge the winner.

What it does

AI Debate Arena turns any topic into a live, voiced debate between two AI personas with opposite personalities. KAI & NATALIE KAI focuses on positive metaphors and can be borderline delusional due to his experiences. NATALIE is factual with an assertive, calm, observant demeanor that shows no emotion.

How I built it

Prompt Engineered ideas with Gemini & Claude AI Chat Agents. Had Gemini & Claude agree on one idea to build towards. Utilized Claude AI in Google AI Studio and during deployment for debugging. Built the app in Google AI Studio and added personal customization.

Gemini 2.0 Flash API is the core engine. When a user submits a topic, the app sends a prompt that defines KAI & NATALIE's distinct personalities, requests a structured JSON response with 6 arguments, and allows the user to pick a winner at or during the debate.

ElevenLabs text-to-speech brings both personas to life with voices that match their personalities. Arguments are sent to the ElevenLabs API individually and played back sequentially via a chain tied to the audio element's onended event. This causes each voice to wait for the previous argument to finish before starting the next.

Challenges I ran into

Issues with customizing the personalities and voices caused the app to crash. KAI's voice broke, and his message would show, but he would not speak. NATALIE would continue speaking regardless of whether KAI spoke. The error was mainly caused by attempting to switch to a premium voice not included in the ElevenLabs trial. Reverted to the original and changed the personalities only. During deployment, encountered problems with getting the app to initialize the build. Utilized screenshots of the issue w/ Claude to debug and get the app running. An ElevenLabs API key is required for the voice interaction to function. Had issues moving to the backend and having it automatically applied when the app loads. Left as is for demo product.

Accomplishments that I am proud of

Deployed a live full-stack application to production. Developing a functional, voice AI application solo in under 12 hours as a first-year AI student. The app works end-to-end with a real-time voice generated for a debate topic. Vibe-coding my first project that works, which mainly required prompt engineering.

What I learned

Product design and app customization can add much more character and depth to its functionality. Adding detail and creativity to the prompt process can change something from boring to entertaining. Also have access to a lot more resources and tools to help me on my journey.

What's next for AI Debate Arena

Could develop it further to include a cleaner UI layout, multiple personalities to choose from, and a leaderboard for each personality. Shareable debate links so users can match with friends and vote together.

Built With

Share this project:

Updates