Snapplot

In game photo of the winner of the scavenger hunter round and AI roast

Inspiration

We drew our inspiration from the chaotic energy of party games and the thrill of treasure hunts. We wanted to move away from the idea of AI as just a rigid technical tool and explore its potential as a personality-driven participant. Our goal was to create a social, unpredictable multiplayer experience where the AI evolves into a creative judge that roasts players, adding a layer of humor and spontaneity to the game.

What it does

Snapplot is a real-time multiplayer mobile app that turns your surroundings into a scavenger hunt arena.

Easy Entry: Players create or join rooms instantly using PIN codes.

The Loop: In each round, the AI generates a unique theme and criteria. Players must race to find and photograph a target object in their physical environment that matches the prompt.

The Verdict: Photos are captured in real-time and sent to the backend. Instead of a boring score, a multi-agent AI judging system evaluates the submissions.

The Roast: A final AI judge selects the winner who best met the criteria and delivers humorous, game-like feedback, roasting the submissions to keep the energy high.

How we built it

We architected a system capable of handling fast-paced, real-time interactions:

Multi-Agent System: We designed a backend where multiple AI judges evaluate each photo submission in parallel to ensure fairness and creativity.

Orchestration: A final "Chief Judge" agent synthesizes these evaluations to select a winner and generate the comedic commentary.

Real-Time Sync: We focused heavily on managing multiplayer game states to ensure that prompts, image uploads, and results were synchronized instantly across all client devices.

Challenges we ran into

The Pivot from Logic to Humor: Initially, we planned to rank entries strictly based on relevance using numerical scores. However, gameplay testing revealed this killed the "party" vibe. We had to pivot, abandoning strict ranking for a system where the AI picks a winner based on "vibes" and humor.

Prompt Engineering: Improving the AI prompts was a significant hurdle. Generating themes that were fun (not just logical) and feedback that was genuinely funny required extensive experimentation and iteration.

Accomplishments that we're proud of

Successfully pivoting the core mechanic: We are proud that we recognized the flaw in our initial "strict scoring" design and successfully shifted to a humor-based model that significantly improved user enjoyment.

Functional Multi-Agent System: getting multiple AI agents to collaborate on judging images in near real-time is a technical feat we are excited about.

Bridging the Physical and Digital: We successfully built an app that forces players to interact with the real world, rather than just staring at a screen.

What we learned

AI Orchestration: We gained deep technical knowledge on designing multi-agent systems and integrating image-based evaluation into fast game cycles.

UX is King: We learned that in party games, timing, feedback, and humor are far more important than technical precision.

Iterative Design: This project reinforced the importance of testing. Without early gameplay tests, we might have stuck with the boring scoring system and missed the fun factor entirely.

What's next for Snapplot

Kid-Friendly "Riddle" Mode: We plan to introduce a gentle mode designed specifically for children, combined with an engaging story where the key plots are directed by the riddles. In this version, the AI will generate riddles instead of standard themes and swap the "roasting" personality for a supportive, encouraging judge.

Social Sharing: Implementing features to easily share the funniest "roasts" and photos to social media to help the game go viral.

Optimization: Further reducing the latency of the AI judging process to make the game even snappier.