ZERONE
Back to projects
Fintech · B2B · Risk Management · 2026

Case 04 — B2B odds compilation & risk platform

Enterprise risk management for online bookmakers: ML-driven odds compilation, sharp-bettor detection, arbitrage prevention and dynamic book balancing in real time.

1000+Events in parallel
M+Bets per period
XGBML fair odds
WSReal-time push

The challenge

A B2B provider for online bookmakers needed a platform that scores risk in milliseconds: multi-source odds aggregation, sharp-bettor pattern detection, internal and cross-bookmaker arbitrage, dynamic margin adjustment. Scalable to thousands of events and millions of bets.

Architecture

A FastAPI core with a WebSocket layer for live push, OAuth2 + JWT auth. XGBoost models derive fair odds from historical data and market consensus. A sharp-detection engine profiles bettors. PostgreSQL primary with read replicas, Redis as hot cache, Prometheus + Grafana for monitoring.

FastAPI · JWT · OAuth2 · SwaggerAPIORCHESTRATORWebSocket layer · real-time pushWSWORKEROdds aggregation · margin engineODDWORKERXGBoost · feature pipeline · retrainingMLWORKERSharp detection · risk scoringSHPWORKERPostgreSQL primary + read replicasDBPRIMARYPrometheus · Grafana · alertsMONWORKER
APIFastAPI · JWT · OAuth2 · Swagger
WSWebSocket layer · real-time push
ODDOdds aggregation · margin engine
MLXGBoost · feature pipeline · retraining
SHPSharp detection · risk scoring
DBPostgreSQL primary + read replicas
MONPrometheus · Grafana · alerts

Pipeline

Odds-compilation flow (sub-second)

  1. 01Multi-source odds aggregated with confidence weights
  2. 02XGBoost model yields fair odds conditioned on historical events
  3. 03Margin engine sets a dynamic margin based on the risk profile
  4. 04Sharp detector flags suspicious bettors in real time
  5. 05Book balancer spreads liability, WebSocket layer pushes to all clients

Technology stack

Python 3.11+FastAPI · UvicornXGBoostPostgreSQL · Read-ReplicasRedisWebSocketsJWT · OAuth2Prometheus · GrafanaOpenAPI 3.0Docker Composepytest · codecov

Outcome

ML fair odds replace manual odds compilation. Sharp-bettor detection significantly reduces exposed liability. WebSocket push delivers price changes sub-second to the frontend and partner APIs. CI pipeline with coverage gate — every new odds strategy is tested before go-live.

Similar challenge?

Talk to us — we listen first, deliver second.

Request a project