Inspiration
I kept seeing the same struggle around me — smart people who knew their stuff but fell apart in high-pressure conversations. Job interviews, visa interviews, debates. The problem wasn't knowledge. It was practice.
Practicing these things is hard. You need another person, you need to schedule time, and honest feedback is rare. I wanted to fix that. The moment I discovered MeDo could generate full-stack applications from natural language, I knew exactly what to build.
What It Does
DebateArena is an AI-powered platform that helps users master two of life's most critical skills — debating and interviewing.
🥊 Debate Mode Go head-to-head against an AI opponent in 5-round debates. Pick any topic — or let AI suggest today's hottest trending topics in real time. Choose your side, argue your case, and get scored by an AI judge on Logic, Evidence, and Persuasiveness after every round. After 5 rounds — a final verdict and personalized improvement tips.
🎙️ Interview Mode Practice realistic interviews with an AI interviewer. Choose from:
- Job Interviews (Google, Amazon, Meta and more)
- College Admissions
- Visa Interviews (F1, H1B, tourist)
- Scholarship Interviews
- Custom scenarios
The AI introduces itself, asks 6-8 tailored questions, probes your answers with intelligent follow-ups, and delivers a detailed feedback report with scores, per-answer analysis, and model answers.
🎤 Voice Input Users can speak their answers out loud via real-time Deepgram speech-to-text — making practice feel like a real interview or debate.
🏆 Gamification 5-tier ranking system (Novice → Grand Master), achievement badges, daily streaks, and a public leaderboard keep users coming back.
How We Built It
This entire platform was built through natural language conversation with MeDo. No manual coding. Here's how we used MeDo's core features:
Full-Stack Generation Described the app in plain English and MeDo generated the complete stack: React + TypeScript frontend with Tailwind CSS and shadcn/ui, Supabase PostgreSQL database with migrations and row-level security, Deno edge functions for all AI processing, and a full authentication system with user profiles.
Multi-Turn Chat The AI opponent and interviewer maintain full conversation history across all rounds. Every response references what was said before — the debate AI counters your specific arguments, and the interviewer asks follow-ups based on your actual answers. This creates genuinely realistic, dynamic sessions.
Plugins & API Integrations
- Groq API (Llama 3.3 70B) — ultra-fast AI responses for debates and interviews
- Deepgram API — real-time speech-to-text via WebSocket streaming for voice input
- Supabase Edge Functions — serverless backend for all AI processing
- GNews-inspired AI topic generation — Groq generates today's hottest debate topics dynamically on every page load
Rapid Iteration via MeDo Chat Every feature was built and refined through ongoing conversation with MeDo — from the gamification system to the video-call-style interview UI to the voice input integration. When something broke, we described the issue to MeDo and iterated. The entire product went from idea to deployed app in under a week.
Challenges We Ran Into
Audio encoding mismatch — Deepgram's WebSocket kept dropping
transcripts because MediaRecorder was sending audio/webm while the
connection expected linear16. Fixed by removing the forced mimeType
and letting the browser auto-detect encoding.
Font configuration error — MeDo used a custom font-montserrat
Tailwind class that wasn't registered in the config, causing a PostCSS
build error that blocked the entire app from loading.
Layout overlapping on mobile — The interview room video panels were clipping the interviewer avatar and question text. Fixed with precise flexbox min-height and overflow adjustments.
API key management — Migrating from OpenAI to Groq required updating multiple edge functions and Supabase secrets simultaneously without breaking the live deployed app.
CORS issues with edge functions — Custom headers were blocked by Supabase's CORS policy, requiring us to move the Groq trending topics call directly to the frontend instead.
Accomplishments That We're Proud Of
- Built a complete full-stack production app entirely through natural language with MeDo — no manual code
- Real-time voice input via Deepgram WebSocket that actually works and feels natural
- Dual mode platform — two completely different AI interaction patterns (debate vs interview) sharing the same infrastructure
- Company-specific interview prep — the AI adapts its questions based on whether you're interviewing at Google, Amazon, McKinsey, and more
- Live trending topics — Groq generates fresh, current, controversial debate topics on every page load
- A gamification system that makes practice genuinely addictive — ranks, badges, streaks, leaderboard
- Going from zero to deployed full-stack app in under a week using only MeDo
What We Learned
- MeDo is genuinely capable of generating production-quality full-stack applications from natural language descriptions
- The key to great MeDo results is treating it like a product manager conversation — clear requirements, one feature at a time, iterative refinement
- Real-time WebSocket APIs like Deepgram can be integrated into MeDo-generated apps with the right prompting strategy
- Splitting complex prompts into focused phases produces far better results than dumping everything at once
- Gamification dramatically increases engagement — the ranking system and badges made testing genuinely fun
What's Next for DebateArena
- 🎯 Multiplayer debates — real-time human vs human debates with live scoring
- 🌍 Multi-language support — interview prep for non-English speakers
- 📹 Video analysis — record yourself and get AI feedback on body language and eye contact
- 🏫 Institutional version — for schools, universities, and corporate training programs
- 📊 Advanced analytics — detailed improvement tracking over weeks and months
- 🤝 Mock interview marketplace — connect with human coaches for premium sessions
For testing use username: test1 password: test123
Built With
- deepgram
- deno
- elevenlabs
- framer-motion
- groq
- html2canvas
- llama-3.3-70b
- medo
- postgresql
- react
- react-router
- shadcn/ui
- supabase
- tailwind-css
- typescript
- vite
- websocket
- webspeechapi
Log in or sign up for Devpost to join the conversation.