MetaTerminal ingests live quotes from all major platforms every 10 seconds, normalizes them into a unified data model, and continuously computes spreads, alerts, and embedding vectors.
The AI layer runs k-means clustering over embedded market titles to group semantically related events — revealing correlated themes invisible in raw data. On-demand gRPC calls to a Python agent generate Claude-powered narratives per market or cluster in seconds.
The trading layer routes smart orders across multiple markets, optimizing for spread savings and supporting both simulated execution and live Solana devnet broadcast mode.
Rails 8.1 Web App (Hotwire + Stimulus) ├── Dashboard, Clusters, Trading ├── Action Cable: /cable (live feed) └── API: /api/v1/* Solid Queue Workers (PostgreSQL-backed) ├── ingest: quote fetch + spreads ├── enrichment: embeddings └── analytics: clustering + AI Python gRPC Agent (port 50052) ├── MarketInsightService ├── ClusterInsightService └── PricePredictionService Data Sources ├── Polymarket (Gamma + CLOB API) ├── Kalshi (REST API) └── Manifold (REST API) On-Chain └── Solana (devnet + mainnet)