Grin Gaming
A real-time odds-pricing engine using high-frequency trading techniques and machine learning to price thousands of micro markets reliably and at scale.
The Challenge
Pricing thousands of fast-moving micro markets by hand is impossible, and stale prices are expensive. The business needed models that re-price continuously as events unfold, with latency low enough to stay ahead of the market and reliability high enough to trade on.
The Solution
We engineered a streaming pricing platform: event data flows through Kafka into feature pipelines, TensorFlow models produce prices, and a FastAPI service backed by Redis publishes them within a sub-40ms budget.
Key Outcomes
- Pricing Latency
- <40ms
- Markets Priced
- 10K+
- Model Uptime
- 99.99%
Technology Stack
- Python
- FastAPI
- TensorFlow
- Apache Kafka
- Redis
What We Built
Streaming data backbone
Kafka ingestion and feature computation that keeps model inputs current as live events change second by second.
ML pricing models
TensorFlow models trained per market family, versioned and shadow-tested before taking live traffic.
Low-latency serving
FastAPI and Redis serving layer engineered against a strict sub-40ms end-to-end latency budget.
Reliability and observability
Health checks, drift monitoring, automatic failover and alerting supporting 99.99% model uptime.
Scale to 10K+ markets
Horizontal partitioning so market count grows without degrading pricing latency.
Impact
The platform prices over 10,000 markets continuously with sub-40ms latency and 99.99% uptime, turning pricing from a manual bottleneck into an automated, always-on capability.
- AI/ML development
- Real-time systems
- Predictive analytics
- Data engineering
- High-performance backend systems
- AI-powered automation
More work
If something here resembles your own situation.
Applications are short, read personally, and answered either way.