Background
Sometimes debating politics can feel like sparring — it’s an art that needs practice, but finding the right partner isn’t always easy. DebateSim is my attempt to create a personal debate partner powered by AI — one you can argue with safely to sharpen your reasoning skills.
The Idea
I wanted to create an Android application that works like a phone call — except instead of chatting with another person, you’re debating with a large language model (LLM).
Each conversation is structured, timed, and documented, giving you a transcript of the debate afterward.
The goal is to practice handling common political arguments, learning to stay concise, logical, and calm under pressure.
It all starts with a question, like:
“Should states be responsible for the school system, or should it be a federal responsibility?”
You pick a side — for or against — and the LLM automatically takes the opposite.
Each turn is limited to 200 words or less, encouraging you to express your position clearly and succinctly. Anything longer is flagged and noted.
Architecture
The first version is built with:
- React Native using Expo (cross-platform for Android & iOS)
- Node.js backend
- Anthropic’s Claude as the debating LLM (though the architecture supports other models too)
This setup allows fast prototyping, voice integration, and easy deployment across mobile platforms.
MVP User Stories
Topic & Side Selection
- As a user, I want to start a debate by entering a question.
- As a user, I want to choose which side (for or against) I will argue.
- As a user, I want the LLM to automatically take the opposite side.
- As a user, I want to start the debate quickly after selecting my topic and stance.
Structured Turn-Taking Debate
- As a user, I want to take turns speaking with the LLM.
- As a user, I want to see when it’s my turn to respond.
- As a user, I want to keep responses under 200 words, with a visible counter.
- As a user, I want the system to flag longer responses.
- As a user, I want to end the debate at any time.
- As a user, I want to speak my argument aloud, and hear the AI’s reply spoken back to me.
Transcript & Documentation
- As a user, I want to see the debate transcript in real time.
- As a user, I want to know who said what and when (with timestamps).
- As a user, I want to review and save the transcript automatically after the debate ends.
What I Learned
Coding it with ClaudeCode was an unexpectedly smooth experience.
I fed in the user stories, walked away for ten minutes, and came back to a ready pull request. From there, I used Cursor to refine the implementation.
However, the constant back-and-forth via pull requests broke my concentration — I missed being “in the loop” while coding. Next time, I’ll stay closer to the active iteration.
A few other takeaways:
- 200 words was too long. The debates dragged on. I switched to a “1 tweet” rule for brevity and energy.
- The app already feels surprisingly real — hearing an AI argue with you while you’re driving is both entertaining and slightly unsettling.
What’s Next
Future improvements:
- Introduce logical fallacies so the AI can argue badly — helping users learn to spot poor reasoning.
- Enable auto-mic activation to keep the conversation flowing hands-free.
- Explore a “car mode” version for debate practice during commutes.