OMX Helsinki — S&P 500 — DAX — NASDAQ 100 — STOXX 600 — EUR/USD — EUR/SEK — BTC/USD — ETH/USD — Euribor 3M — Euribor 12M —
AI Engineer

AI Engineer

Technology and AI 889,000 subscribers

Do Not Let Models Invent Numbers – Vinoo Ganesh, Kepler

25.10.2026 8,894 views
Published: 25.10.2026 Category: Technology and AI Views: 8,894

Episode details

Episode description

In finance, you never show a figure you can't trace. Vinoo Ganesh, CEO of Kepler (previously Palantir and Citadel), explains Kepler's architecture where models cite numbers but never write them: retrieval finds the source, deterministic code does the math, and a post-processing filter strips any figure that can't be verified. He covers the citation registry built from more than 30 million SEC documents, hybrid search and re-ranking on turbopuffer, tenant isolation, and why plain RAG produced thousands of wrong numbers in their evals. Chapters 00:00 Intro 01:47 Why LLMs and financial numbers don't mix 04:02 Models cite numbers, code does the math 04:30 Building a citation registry from SEC filings 06:03 How Kepler answers a question end to end 08:27 Structuring facts: value, unit, scale, period 09:48 Hybrid search and re-ranking on turbopuffer 11:30 Tenant isolation and security 12:37 Making every number traceable 14:42 Stripping what can't be verified 16:41 What plain RAG gets wrong 18:32 Kepler in action 19:46 Where the guarantees end Vinoo Ganesh, Kepler Join us in SF for the second annual AIE CODE summit: https://aie.codes our last flagship summit of 2026. Applications live now!

Open the AI assistant chat. The chat loads only when you open it.