May 3, 2026
Audit the bot, baby
How Kepler built verifiable AI for financial services with Claude
Wall Street wants AI it can prove, but the comments section is not buying the hype
TLDR: Kepler says it built an AI finance tool that can show exactly where every answer came from, a huge deal in a business where mistakes can be costly. Commenters were split between calling it a smart safety layer and saying it proves the chatbot is the weakest part of the whole system.
Kepler’s big pitch is simple: let people ask finance questions in plain English, then make sure every number can be traced back to the original documents. The startup says it pulled together a mountain of records in record time and built a system where Claude, Anthropic’s artificial intelligence chatbot, does the interpreting while old-school code handles the actual digging and math. In a world where a bad number in a bank report can cause serious trouble, that sounds like catnip for nervous finance firms.
But the real fireworks were in the comments. On Hacker News, one camp basically said, “Wait... so the good part is the non-AI part?” The loudest skeptics mocked the whole setup as proof that chatbots are just extra hassle wrapped in fancy branding. One commenter sneered that it was “very encouraging to see plain old deterministic infra” and then twisted the knife: maybe everyone would be better off without the language model at all.
Others defended the idea, arguing that the chatbot’s real job is to translate messy human questions into clean steps that reliable software can follow. But then came the credibility bombs: one user dragged the founders’ Palantir past into the thread, while another complained the title oversold the product because 94% accuracy is nowhere near good enough when real money and regulations are on the line. In other words: Kepler tried to sell “trust,” and the internet responded with the oldest audit of all — vibes, suspicion, and a lot of side-eye.
Key Points
- •Kepler built a deterministic trust and verification layer for AI in financial services and uses Claude as the reasoning layer in its platform.
- •The company says it indexed 26M+ SEC filings, 50M+ public documents, 1M+ private documents, and 14,000+ companies across 27 global markets in less than three months.
- •Kepler’s founders, Vinoo Ganesh and John McRaven, previously worked at Palantir and spoke with 147 financial firms before founding the company.
- •The article says financial analysis requires auditable, source-verifiable outputs, which traditional tools and end-to-end AI systems do not fully provide on their own.
- •Kepler reports that Claude outperformed other frontier models on long, multi-step financial tasks and was better at surfacing ambiguity for human clarification.