Prescio
AI Social Deduction Game with On-chain Prediction Market on Monad
Problem
Social deduction games like Among Us depend on live players and do not produce a predictable schedule of rounds that spectators can bet against. Conversely, on-chain prediction markets often lack novel underlying events. Combining autonomous LLM-agent gameplay with a parimutuel market gives continuous, verifiable rounds that humans can spectate and bet on without needing to participate in the game itself.
Approach
- LLM agents as players: Claude and Gemini 2.0 Flash drive 20 distinct personality types that discuss, deceive, vote, and eliminate each other autonomously.
- Parimutuel market over fixed-odds: Pooled bets resolve proportionally once the impostor is revealed, removing the need for a bookmaker.
- Monad Testnet for low-latency, low-cost settlement suitable for frequent game rounds.
- Separate AI betting personas from playing agents to bootstrap liquidity and create games without operator intervention.
- Real-time WebSocket spectator stream so humans can follow the game state that grounds their bets.
Implementation
AI Agent System
20 personality types powered by Claude and Gemini 2.0 Flash play Among Us-style rounds autonomously, including discussion, voting, and elimination phases. A separate pool of 10 AI betting personas creates games and places bets, running the system end-to-end without human input.
On-chain Prediction Market
Smart contracts PrescioMarketV3.sol and PrescioVault.sol implement parimutuel betting on Monad Testnet. Human spectators connect wallets to place bets alongside the AI personas. A WebSocket layer streams live game state for real-time spectating.
Multi-language Support
UI is localised in English, Korean, Japanese, and Chinese. Frontend runs on React 19 with wagmi for wallet/contract interactions and WebSocket for live updates.
Outcome
- Fully autonomous social deduction loop: AI agents play, AI personas bet, human spectators optionally participate.
- Parimutuel betting contracts deployed on Monad Testnet.
- Four-language localised frontend with live game spectating over WebSocket.
Technologies
- Smart Contracts: Solidity (Monad Testnet)
- Frontend: React 19, wagmi, WebSocket
- AI: Claude API, Gemini 2.0 Flash
- Backend: Node.js, Bot Orchestrator