SympactLab
Multiplayer Mystery Game Platform — SympactAI Edition 2
Video Demonstration
System Engineering Architecture
Designed & Engineered by Mazen Temani
End-to-end intelligent engineering pipeline built with clean separation of concerns, semantic data retrieval, and robust asynchronous APIs.
The integration week needed a team-building engine, not just a game — something that forced real collaboration between the AI/ML teams while staying fair and cheat-proof at scale.
Built a live multiplayer mystery with a server-authoritative FSM for game phases, per-team asymmetric clue distribution, private group chats and notes boards, and cross-group channels — contract-first (Orval/Zod) with a WebSocket + React Query polling hybrid and GCS presigned proof uploads.
Ran the full SympactAI Edition 2 game end to end — turning cross-team collaboration into the core game mechanic, with real colleagues cast as story suspects.
Technical Deep Dive
A live multiplayer murder-mystery game platform built for the integration week of SympactAI's second edition (TAIS NextGen). Teams join with a session code, pick a predefined team name, and investigate 'The SympactLab Incident' — a fictional AI-fraud storyline (sabotaged model, inflated accuracy) casting real colleagues as suspects. Evidence is deliberately split so no single team can solve the case: collaboration becomes a game mechanic, not a feature. Groups get private chats and notes boards for deductions, plus cross-group channels to share — or strategically mislead rivals with — exclusive clues. Built with a server-authoritative FSM for game phases, a WebSocket + React Query polling hybrid, per-team asymmetric clue distribution, an Orval/Zod contract-first stack with generated clients and schemas, Drizzle ORM over PostgreSQL, and GCS presigned uploads for proof submissions. The demo shows the Director Console running the 'Bayesian Federated Learning for Bank Fraud Detection' case from investigation to deliberation.