Kepler, founded in 2025 by former Palantir engineers Vinoo Ganesh and John McRaven, built a trust and verification layer for AI in financial services. After speaking with 147 financial firms who all expressed the same problem—everyone wanted AI for research but nobody trusted the output—the duo created deterministic infrastructure that lets analysts ask questions in plain English and receive instantly verifiable answers.

The team benchmarked across all frontier models and found that while straightforward queries performed comparably, only Claude maintained accuracy on long, multi-step plans. Other models started taking shortcuts by the fourth or fifth step. Most importantly, Claude stops and asks for clarification when terminology is ambiguous, while other models guess and keep going—a critical feature when one wrong assumption breaks everything downstream.

The engineering approach combines Claude's reasoning with deterministic execution environments for provably correct computations. The team built a proprietary ontology mapping financial concepts to precise formulas, deployed Opus 4.7 for complex reasoning and Sonnet 4.6 for higher-throughput stages, and trained specialized models scoring 94% accuracy on financial statement mapping versus 38-46% for other models.

Kepler has indexed 26M+ SEC filings across 14,000+ companies in 27 global markets, with every number traceable back to its exact source document. The separation between AI reasoning and deterministic infrastructure lets a small team build at startup speed while maintaining the audit trails financial institutions require. The platform is SOC 2 Type II certified.