Case Study: Fal achieves millisecond search across 300M+ media assets with Turbopuffer

Fal scales media search to 300M+ assets with Turbopuffer

fal, a generative media platform for developers, faced the challenge of building a search function for its Assets product that could scale to hundreds of millions of media files. They needed a solution that could handle hybrid search, large inclusion filters, and strict per-tenant data isolation without compromising on latency or recall for their users.

Using Turbopuffer, fal implemented a filtered hybrid search system that combines BM25 text search and vector ANN over media embeddings. The solution successfully indexes over 300 million assets across 225,000+ isolated namespaces, achieving a p50 latency of just 17ms and enabling immediate, secure search across their entire user base.

Fal