About Vinoo Ganesh
Vinoo Ganesh is CEO and co-founder of Kepler, a company building verifiable AI for financial services. He was previously head of business engineering at Citadel, working across the firm on the software its portfolio managers used to generate alpha, and spent several years at Palantir, where he led Spark work and built a forward deployed engineering program.
In an AI Engineer talk, he explains why Wall Street still doesn't trust AI and how Kepler is architected so that a model structurally cannot make up a number: code assigns a pointer to every fact during preprocessing, the model acts as a planning analyst, deterministic code does retrieval and arithmetic, and any number the model cannot dereference is refused.
Notable Positions
On Hallucination
Hallucination is structural, not a bug better models will remove. "Another word for hallucination is alpha": the creativity that makes analysts valuable is the same mechanism that invents numbers. See Hallucination.
On Verification
Verification cannot be tacked on at the end. Wrong numbers come from retrieving the wrong document, pulling the wrong number from the right document, or mental math; each needs a structural fix, and the final post-processing check only catches what's left. See Grounding.
On Finance AI Adoption
Most finance AI tools are being used to format PowerPoint slides, and nobody has built a DCF end to end. Firms will adopt the first tool whose numbers they can click on and trace.
Key Quotes
- "Code writes every number." (on Kepler's architecture)
- "A lookup cannot get it wrong. Search can." (on deterministic retrieval)
- "We're now out of the search phase of AI. We need to be building infrastructure." (on what builders should do next)
Related Reading
- Grounding - Kepler's approach as an architectural constraint
- Hallucination - The problem Kepler is built around