GameScript streams AI predictions and analyst insight to every fan with sub-100 ms latency
GameScript is a sports analytics platform that fuses AI predictions with human analysis and streams the result to fans as the game plays out. Ingestion, querying and in-database inference run on SurrealDB's multi-model engine.
Challenge
Sports data changes moment by moment. Odds shift with every play, and a model of the game has to move with them rather than being fixed at kickoff.
Live games bring millions of concurrent events and massive write volumes. On top of that, AI predictions and analysts' annotations have to land in one view at the same instant, because an insight about a play is worthless a minute later.
Building this on several databases, one for documents, one for vectors, one for relationships, would have slowed development and multiplied the operational load for a team that needed to ship quickly.
Solution
GameScript modelled users, AI insights and annotations in SurrealDB's one multi-model engine. Vector, document and graph data live in the same store, so the platform has one system to build against and to run.
SurrealQL blends joins, graph traversals and live queries, which is how predictions and annotations reach fans as they arrive. SurrealDB Studio and the embeddable Rust binary sped up prototyping and the rollout that followed.
Machine learning inference runs inside the database, so a prediction is made and served from the engine at sub-100 millisecond latency, at the volumes a live game produces.
Results
A fast launch
A production-grade platform was delivered in record time, on one engine from prototype through to production.
Real-time at scale
Millions of writes during live games with sub-100 millisecond latency.
A single system
Vector, document and graph data in one store, with one query language over all of it.
Flexible schema deployments
New sports markets roll out with minimal overhead, because the schema bends instead of breaking.
