DissCourse

Inspiration We wanted to explore what happens when you put history's most iconic minds in the same room. What would Socrates say to Elon Musk? Would Lincoln and Trump agree on anything? The hackathon theme of cloning gave us the perfect angle — we could clone the personalities, speech patterns, and worldviews of real figures and let them clash.
What it does DissCourse lets you pick two AI clones of real historical and contemporary figures, give them a debate topic, and watch them argue it out in real time. Each clone speaks in character using their actual beliefs and rhetorical style. You can inject questions from the audience mid-debate, and after 3 rounds each you vote on who made the better argument.
How we built it Frontend: React + TypeScript + Tailwind CSS (Vite) Backend: Python HTTP server with Ollama running llama3.2:3b locally for fast inference Voice: Browser Web Speech API with per-character pitch and rate profiles Character prompts: Carefully crafted system prompts for each figure with historical knowledge cutoffs.
Challenges we ran into Getting the AI to stay in character and on topic without going off on tangents Enforcing strict historical knowledge boundaries Browser autoplay policies blocking audio Keeping responses concise. The model naturally wanted to write essays Accomplishments that we're proud of Each character genuinely sounds and argues like themselves The debate flow feels natural. Real back and forth, not just monologues Historical figures stay within their era's knowledge The whole thing runs locally with no paid API required
What we learned Prompt engineering is everything. Small wording changes dramatically affect character authenticity Local LLMs are surprisingly capable for structured roleplay tasks Browser TTS is more flexible than expected with pitch/rate tuning per character What's next for DissCourse Add more characters (MLK, Einstein, Taylor Swift, etc.) Let users create custom debaters with their own personality descriptions Multiplayer mode where two users each control a debater A scoring system that analyzes argument quality, not just user votes