Debatium

Debatium — AI-powered debate platform to sharpen critical thinking. Spar against smart AI or friends, in English & Vietnamese, with real-time voice responses. Practice anytime, anywhere.

Comment

Inspiration

Debate is one of the most powerful tools for developing critical thinking, public speaking, and logical reasoning - yet access to quality debate practice is still limited to well-resourced schools and clubs. Most students who want to improve have no opponent to practice with, no coach to give feedback, and no platform built specifically for their needs.

We were inspired by the gap between the demand for debate training and the tools available. As university students ourselves, we've experienced firsthand how difficult it is to find sparring partners at all hours, especially across language barriers. We wanted to build something that democratizes debate practice - an always-available, AI-powered debating partner that challenges you, adapts to your level, and speaks back to you in your own language.

Debatium was built to fix both problems at once: give everyone a relentless, intelligent opponent that commits fully to a position, and a judge who evaluates performance with the rigor of a trained coach, not the politeness of a chatbot.

What it does

Debatium is an AI-powered debate training platform with two core systems working in tandem:

  • Debate an AI opponent - Face off against an intelligent AI that argues the opposing side of any topic, powered by advanced language models. The AI constructs structured arguments, rebuttals, and counterpoints in real time.
  • Multilingual support - Debate in English or Vietnamese, making it accessible to a wider Southeast Asian audience (powered by Valsea).
  • Text-to-Speech (TTS) - Hear the AI's arguments spoken aloud, creating an immersive, speech-like debate experience that mirrors real competition.
  • Human vs. Human Sparring - Create a private spar room and invite friends to debate head-to-head - no AI involved. A pure human debate experience with session management built in.
  • Collaborative AI Sparring - Invite friends to join a shared session and practice together against the AI as a team.
  • Session History & Calendar - Track all past debates, review your performance history, and manage upcoming debate sessions through a built-in calendar.

How we built it

Debatium was built with a modern, production-grade full-stack architecture:

  • Frontend & Framework: SvelteKit with Svelte 5, delivering reactive, component-based UI with excellent performance.
  • Language: TypeScript throughout for type safety and maintainability.
  • Database: Neon (serverless PostgreSQL) managed with Drizzle ORM for schema-first, type-safe database access.
  • Authentication: Better-Auth for secure, session-based user authentication.
  • AI Integration: Vercel AI SDK orchestrating calls to OpenAI for argumentation logic, and Valsea for Vietnamese language support and Text-to-Speech generation.
  • Styling: Tailwind CSS 4 with Bits UI and Lucide Svelte for a polished, accessible, and responsive interface.
  • Deployment: Live and deployed at debatium.mncuchiinhuttt.dev.

Challenges we ran into

  • Maintaining debate coherence - Getting the AI to track the full argument history across multiple turns and respond in a contextually relevant, logically structured way required careful prompt engineering and context management.
  • Making the AI stop being nice - By default, AI models are tuned to be agreeable and avoid conflict - the exact opposite of a good debate opponent. We had to aggressively prompt-engineer the AI to actually argue back, challenge weak points, and hold its position under pressure without caving to the user's arguments.
  • Multilingual AI quality - Ensuring the Vietnamese debate experience was as natural and nuanced as the English one was a significant challenge. Integrating Valsea alongside OpenAI required custom bridging logic.
  • Real-time TTS latency - Streaming AI-generated text to speech in real time while keeping the UI responsive required careful handling of async data streams and buffering.
  • Collaborative session synchronization - Building a private "spar" feature where two users share a live debate session introduced state management complexity, especially around turn-taking and session ownership.
  • Schema design under time pressure - Designing a flexible database schema with Drizzle ORM that supports users, sessions, debate history, and calendar events - all within a hackathon timeframe - required fast, decisive architectural choices.

Accomplishments that we're proud of

  • Deployed a production-ready app accessible live at debatium.mncuchiinhuttt.dev — fully functional end-to-end.
  • Multilingual AI debate - One of very few debate platforms that natively supports Vietnamese with voice, making it uniquely relevant to the Vietnamese and Southeast Asian student community.
  • Immersive TTS experience - Successfully integrated real-time voice output so the AI literally speaks its arguments back to the user, elevating the experience beyond a simple chatbot.
  • Full-stack in a hackathon - Delivered a complete product with authentication, database, AI, real-time features, and a polished UI - all within the hackathon window.
  • Collaborative spar mode - The private session feature transforms Debatium from a solo tool into a social practice platform.

What we learned

  • Prompt engineering is an art - Crafting system prompts that make the AI behave as a structured debate opponent (with constructive speeches, rebuttals, and cross-examination) took multiple iterations.
  • Drizzle ORM + Neon - This combination proved to be an excellent developer experience for rapid schema iteration with serverless PostgreSQL - something we'll carry into future projects.
  • The value of constraints - A hackathon forces you to prioritize ruthlessly. We learned to ship a focused, working core rather than an overengineered half-product.
  • Accessibility in language - Building for Vietnamese users taught us how important it is to design AI systems with non-English speakers in mind, not as an afterthought.

What's next for Debatium

  • Judging & Scoring - Integrate an AI judge that scores each debate round on argumentation quality, evidence use, and delivery - giving users structured feedback to improve.
  • Debate formats - Support structured formats like British Parliamentary (BP), Lincoln-Douglas (LD), and Asian Parliamentary (AP) with timed rounds and format-specific rules.
  • Community & Leaderboards - Build a public debate arena where users can challenge each other, track rankings, and watch replay transcripts of top debates.
  • Mobile App - Bring Debatium to iOS and Android so users can practice on the go.
  • School & Club Integration - Partner with debate clubs and universities across Vietnam and Southeast Asia to use Debatium as a core training tool.
  • Fine-tuned debate models - Train specialized models on debate transcripts and competitive rounds to produce even sharper, more pedagogically useful AI opponents.

Built With

Share this project:

Updates