R.A.C.C.O.O.N

A multiplayer trivia game that covers everything from your roommates secrets to your bus ticket serial number.

Comment 3

Inspiration:

This project started organically with a group of close friends on just spending time together. During our winter break, we found ourselves constantly coming back to playing trivia and social games to bond. Further, we realized that many multiplayer games like Jackbox are incredibly fun, but they don’t adapt to what you actually care about: your classes, your friendships, or real-world events. At the same time, we found ourselves rushing through assignments just to make time for these games, treating fun and productivity as completely separate parts of our lives.

At the same time, prediction platforms like Polymarket have seen billions in trading volume, proving how engaging it is to forecast outcomes and think probabilistically about the world. What was truly missing is a layer that combines all three and makes it personal. A platform that is fun, real-world based, and social.

What it does

Our platform is a real-time multiplayer trivia and prediction game that blends social gameplay with learning and forecasting. In addition to Polymarket and social data, it takes PDFs of any content and translates them seamlessly onto our themed Raccoon Robbery game with an AI host.

Players join a synchronized game session and compete across four core modes (can also be mixed).

Study Mode Upload lecture notes, problem sets, or PDFs, and instantly transform them into a multiplayer trivia game. This turns studying into an interactive, competitive experience driven by active recall.

Polymarket Mode Answer questions about real-world events, earning points for accurate predictions. This introduces strategy, psychology, and crowd prediction into gameplay. Choose whichever PolyMarket category you prefer!

Custom Mode The most chaotic mode; players answer their own prediction questions about their friends. From test scores to sleep habits to social behavior, this becomes a live social experiment where knowing people is just as important as knowing the facts.

Scrappy R.A.C.C.O.O.N Mode In Scrappy mode, every junk file on your computer becomes worthy of trivia! An old essay? Toss it in. Your bus ticket to YHacks? Your brother definitely should be able to guess the time that you departed.

How we built it

We built this project as a full-stack, real-time multiplayer system with a strong AI layer: Frontend: React, TypeScript, Vite, Tailwind CSS, Framer Motion Backend & Infra: Node.js, Vercel (serverless functions), Supabase (auth, database, real-time sync) AI & Content Generation: Gemini 2.5 Flash for dynamic quiz generation and in-game quips Voice Layer: ElevenLabs for narration and character-driven gameplay Data Integration: Polymarket Gamma API for real-world prediction questions Document Processing: PDF.js for extracting study materials into playable content Design & Workflow: Figma, Veo, Nana Banana, Codex, GitHub, GitHub Actions We leveraged Supabase’s real-time capabilities to synchronize players across sessions, while AI handles content generation, narration, and dynamic game flow. However, we also have strong token-use minimizing mechanisms in place including caching frequently asked questions, phrases, and audio-clips.

Challenges we ran into

One of our biggest challenges was real-time gameplay synchronization. Ensuring that multiple players could join the same game and stay in sync, especially across different devices like phones and laptops required careful handling of state, latency, and event timing. Getting the experience to feel smooth and responsive across platforms took multiple iterations. We also faced an early infrastructure decision between Colyseus and Supabase for multiplayer support. While Colyseus offered a game-server approach, we ultimately pivoted to Supabase’s real-time capabilities and framer compatibility, which better aligned with our stack and allowed us to move faster with integration and deployment. Another challenge was building a seamless pipeline from PDF parsing to question generation read aloud by an AI host. We used PDF.js to extract text, then leveraged Gemini to generate questions, stored them in Supabase, and passed them into ElevenLabs for narration. Coordinating this multi-step pipeline, while maintaining consistency and quality required careful orchestration. Working with AI introduced its own challenges. Both Gemini and ElevenLabs occasionally produced hallucinations or inconsistent outputs, especially in voice narration. We mitigated this by refining prompts, adding structure to generated content, and implementing basic validation layers. We also underwent a major product pivot. Initially, we aimed to build a Jackbox-style study tool, but realized the experience became significantly more engaging when we introduced a prediction layer centered around friends and social behavior. This shift reshaped both our product direction and technical design. Finally, we had to think carefully about cost and scalability. Since our platform relies on APIs like Gemini and ElevenLabs, managing usage efficiently was critical. We implemented ways to reduce unnecessary calls, reuse generated content, and design the system to remain sustainable as usage grows.

Accomplishments that we're proud of

We’re especially proud of how much personality and polish we were able to bring into the product in such a short time. First, our team worked incredibly well together; we had a strong mix of technical and creative skills, which allowed us to move quickly while also building something that feels thoughtful and fun. One of our favorite features is the unhinged AI voice narrator, which adds humor and energy to the game and makes each round feel alive. Paired with this, we designed character archetypes inspired by student identities (Harvard, Yale, Stanford, MIT, etc.), making the experience more relatable and socially engaging.

We’re also proud of our end-to-end AI content pipeline. We built a system that can take raw PDFs and txts—even complex or poorly formatted ones—and extract text, generate structured questions using Gemini, store them in Supabase, and deliver them in real time as part of a multiplayer experience. This effectively allows users to turn any digital content into a live, interactive quiz with friends. We also successfully implemented real-time multiplayer synchronization using Supabase, enabling multiple players to join, respond, and receive results simultaneously with minimal latency. Ensuring this worked smoothly across devices—including mobile—was a major milestone.

We also invested heavily in user experience and immersion. We designed visuals, animations, background music, and sound effects to create a game that is not only interactive, but also visually and auditorily engaging.Overall, we’re proud that we were able to integrate AI generation, voice systems, real-time infrastructure, and creative design into a single platform that feels dynamic, social, and genuinely fun to use—while also pushing the boundaries of how games can incorporate personalization and prediction.

What we learned

Getting Gemini and ElevenLabs to consistently generate usable high quality outputs took real iteration. We had to learn how to guide the models, structure responses, and actually make them fit into a live game environment. Real time multiplayer was also tough. Syncing multiple players across devices involved small delays, mismatched states, and timing issues that completely disrupted the experience. It forced us to think n about whether our game was smooth and enjoyable for everyone playing. One of the coolest moments for us was realizing when our PDFs came alive. This was the result of hours of work, but turning notes into a live multiplayer story-line quiz completely felt magical as our old essays became part of our game– we encourage you to try out! It made us realize that anything digital can become interactive if you build the right system around it. A lot of our biggest improvements did not come from coding, but from conversations. Talking to each other, getting feedback, and even speaking with mentors like the MLH rep pushed us to rethink our direction. That is what led us to pivot from a study tool into something much more social and prediction driven. Building R.A.C.C.O.O.N was loads of fun, and we think that it was because of the fun that made R.A.C.C.O.O.N the goofy, fun, and socially relevant game that it is.

What's next for R.A.C.C.O.O.N

Our next step is to play this game with as many friends and family as we can and get feedback whilst seeing what they think. Perhaps professors would be interested in a tool like this!

Built With

  • codex
  • database
  • elevenlabs
  • figma
  • framer-motion
  • gemini-2.5-flash
  • nana-banana
  • node.js
  • pdf.js
  • polymarket-gamma-api
  • react
  • real-time-sync)
  • supabase-(auth
  • tailwind-css
  • typescript
  • veo
  • vercel-serverless-functions
  • vite
Share this project:

Updates