Case Study: Kepler achieves fast, low-cost search across 100M+ financial documents with Turbopuffer

A Turbopuffer Case Study

Preview of the Kepler Case Study

Kepler searches 100M+ documents with turbopuffer at 10x lower cost

kepler, a company building deterministic AI infrastructure for finance, needed to search a massive and growing corpus of over 100 million financial documents. Their challenge was to provide verifiable AI responses that trace back to exact sources without incurring the outrageous costs of traditional disk-based search engines, which would require constant capacity planning and added nodes.

turbopuffer solved this by building indexes directly on object storage, dramatically lowering the marginal cost of adding each new document. Using turbopuffer's hybrid vector and full-text search within isolated namespaces, kepler achieved 85ms p90 latency and 10x lower cost. This allowed them to search their entire long-tail corpus and keep their "always right" promise with bounded latencies for their verifiable AI outputs.


View this case study…

Turbopuffer

13 Case Studies