Prediction-market intelligence · 2026 · Data ingestion, backend architecture, signal modeling, interface, deployment
Smart Money Tracker
A bounded current-sample prediction-market monitor that turns two incompatible public feeds into one inspectable signal surface.
The problem
Kalshi and Polymarket expose useful public activity through different payloads, identifiers, price conventions, and participant data. Reviewing those feeds separately makes it difficult to compare prints, identify large trades, cross-reference known leaderboard wallets, or see where multiple smart-money signals align.
Architecture
Key decisions
One bounded model, two incompatible feeds
Kalshi and Polymarket trades are mapped into a shared schema before any display or aggregation logic runs. Platform-specific parsing stays at the ingestion boundary, while the tape and signal panels operate on one contract.
Separate the production demo from persistent analysis
Vercel serves an honest, short-lived public snapshot without claiming durable storage or streaming. The original local worker remains the persistent mode, where background pollers, SQLite deduplication, and SSE are appropriate.
Derive explainable signals instead of predictions
Whale thresholds, observed wallet volume, net market flow, and smart-money agreement are direct transformations of public trade data. The interface explicitly presents them as monitoring evidence, not financial advice or predictive certainty.
Keep the operational surface read-only
The deployed project uses unauthenticated public market endpoints and exposes no order-entry path, account credentials, or trading automation. Input bounds and a short cache protect the serverless boundary.