A San Francisco AI company turning the most paperwork-heavy industries on Earth into something a machine can read - and a regulator can audit.
An opening scene, in three acts.
Somewhere in a glass tower in Manhattan, a junior analyst is reading the 187th page of a commercial lease. It is 11:42 p.m. She has been awake since 6. Her highlighter is dying.
This is the room Kolena was built for. Not the headline AI demos. Not the chatbot you flirted with last Tuesday. The other rooms - the back offices, the loan packages, the claim files, the lease abstracts. The places where high-stakes decisions are still made by people who have stopped feeling their fingers.
Kolena, the company, lives in San Francisco. Kolena, the product, lives in those rooms. It reads the 187 pages. It pulls out the renewal clauses, the escalation provisions, the indemnity language. And then - this is the part that matters - it shows you exactly where it found each one. Underline, page number, paragraph, source.
Most AI tells you the answer. Kolena tells you the answer and hands you the receipt.
Founded in May 2021 by Mohamed Elgendy, Andrew Shi, and Gordon Hart, Kolena started life as something subtly different from what it is today. The original product was a rigorous ML testing platform - a sort of crash-test facility for computer vision, NLP, and large language models. Teams could push their models through, and Kolena would find the failure modes the engineers hadn't.
The pivot, when it came, was less a left turn than a logical conclusion. If you spend your days helping enterprises figure out whether their AI works, you eventually meet the enterprises that don't have AI yet - but desperately need it. The insurers, the banks, the real estate operators, the regulators. The places where a wrong answer isn't an embarrassing screenshot. It's a lawsuit.
Three workflows, one operating system.
Reads commercial and residential leases. Pulls renewal options, escalation clauses, indemnities. Cites every line.
Reviews claim packages end-to-end. Validates evidence. Surfaces inconsistencies before they become disputes.
Processes loan packages and compliance files with the audit trail regulators ask for - and the speed underwriters demand.
The company's first act: end-to-end model testing for CV, NLP, LLMs, and multi-modal. The DNA is still in the product.
Before Kolena, they shipped AI inside Amazon, Palantir, Rakuten, and Synapse.
Author, AI builder, and the company's public face. Spent six years in AI divisions at Amazon and beyond before launching Kolena in May 2021.
The architecture lead. Builds the rails the workflow agents run on. Believes in showing the work.
Product. Translates the rigor of model evaluation into something a non-engineer can use without breaking it.
$21 million, two rounds, one consistent thesis.
| Round | Date | Amount | Lead / Investors |
|---|---|---|---|
| Series A | September 26, 2023 | $15M | Lobby Capital (lead), SignalFire, Bloomberg Beta, 11.2 Capital, Zero Prime Ventures |
| Seed | 2021 | ~$6M | SignalFire, Bloomberg Beta |
The Series A press release, in its own boilerplate language, called this an investment in "a systematic and rigorous AI quality framework along with toolings that empower businesses to build AI systems society can trust." The translation, in plain English: investors are betting that the AI companies that win the enterprise will be the ones who can prove what they did.
There are, generously, ten thousand AI startups. There are perhaps a few dozen building for the corner of the market where a single hallucination ends a career. Kolena is in the latter group.
The company's product positioning is unusual. It does not lead with model size. It does not lead with benchmarks. It leads with reasoning logs, citations, SOC 2 Type II, HIPAA, and an audit trail. These are not features that demo well on Twitter. They are features that get a procurement officer in a regional bank to sign.
That, more or less, is the bet. Most AI buyers will eventually be procurement officers in regional banks. Kolena is just there early.
Talks, demos, and the founder making the case.
It is 11:42 p.m. The lease is 187 pages. The highlighter is, mercifully, no longer dying - because no one is using it. The junior analyst is at home, asleep, where junior analysts are supposed to be at 11:42 p.m.
The lease has been read. The renewal clauses are highlighted, but not in yellow ink. They are tagged in a Kolena dashboard, each with a page number, a paragraph reference, and a confidence score. In the morning, a partner will scan them in eight minutes. She will accept most. She will reject one. She will be able to tell you, line by line, why.
This is the room Kolena was built for. And the small, unflashy promise the company makes - that the AI will tell you what it did, and you will be able to check - turns out to be a much bigger idea than it first sounds. It is, in fact, the thing that lets AI into the rooms where the stakes are real.
The 35 people in San Francisco are not trying to win the chatbot war. They are trying to make AI legible to people whose job is to be skeptical. If they pull it off, the glass tower at 11:42 p.m. will look very different. It already does.