Kepler Builds Verifiable AI for Financial Services With Claude

Summary

Financial services firms have long struggled to trust AI-generated outputs in heavily regulated environments where every figure must be auditable against source documents. Kepler, a startup founded by former Palantir engineers Vinoo Ganesh and John McRaven, has built a platform that combines Claude with deterministic infrastructure to create a "trust and verification layer" for financial research and analysis. The platform indexes over 26 million SEC filings, earnings transcripts, and other financial data across 14,000+ companies in 27 global markets, enabling analysts to ask plain English questions and receive instantly verifiable answers.

During benchmarking across frontier AI models, Kepler found that Claude consistently outperformed competitors on long, multi-step financial analysis tasks. While other models began to drop constraints and take shortcuts by the fourth or fifth step of complex analyses, Claude maintained focus throughout the entire plan. More critically, Claude's approach to uncertainty—stopping to ask for clarification when terms could have multiple meanings—proved more valuable in practice than raw benchmark performance, since a single wrong assumption early in financial analysis invalidates all downstream calculations. By separating the interpretation and reasoning work (handled by Claude) from deterministic verification and computation (handled by Kepler's infrastructure), the team created a system that balances both flexibility and auditability.

• The platform addresses the core barrier to AI adoption in finance: the need to audit and verify AI-generated outputs against original source documents