Quantifind reads the world for a living.

A small company doing the unglamorous work of finding bad money.

Acompliance analyst at a Tier-1 bank logs in on a Monday morning. Her queue has 1,400 alerts. Most are noise - a customer with a name resembling someone on a sanctions list, a news mention that turned out to be a different John Wang. By Friday, she has cleared maybe 200. The bad ones, the true positives, are still in there. Somewhere.

Quantifind exists for that Monday morning. The Palo Alto company - about 95 people, $135 million raised, headquartered at 444 High Street between a cafe and a salon - builds AI that reads the public internet and tells compliance teams where to look. Not which alert is interesting in theory. Which one is interesting today.

It is not glamorous work. There is no consumer app, no viral moment, no chief evangelist on a TED stage. There is, instead, a platform called Graphyte and a quiet customer list that includes the Department of Defense, a digital bank named Varo, an anti-trafficking nonprofit named Polaris, and a handful of banks that prefer not to be named at all.

Compliance is drowning in false positives.

Banks are legally required to know their customers. They are required to monitor transactions. They are required to file Suspicious Activity Reports when something looks off. These obligations are not new. What is new is the volume.

A modern bank screens tens of millions of names against watchlists, sanctions, adverse media, and politically exposed person databases - every single day. The legacy tooling for this work was designed in the 2000s, back when fuzzy string matching was considered cutting edge. The result, predictably, is alert fatigue: somewhere between 95 and 99 percent of alerts at most institutions turn out to be nothing. Analysts spend their careers clearing noise.

Meanwhile, the actual criminals adapt. Shell companies are spun up overnight. Human-trafficking rings move money through unremarkable LLCs. Sanctioned entities re-emerge under new names in foreign-language press. The signal is in the data. The data is just not in any tidy table.

Two physicists, an unexpected pivot, and a stubborn idea.

Ari Tuchman and John Stockton met at Stanford. Both were quantum physicists - Tuchman from Stanford, Stockton from Caltech - the kind of people who think in tensors before breakfast. In 2009 they founded Quantifind on a thesis that was almost charmingly broad: that signals about anything in the world were already lurking inside unstructured text, and that a sufficiently clever model could pull them out.

The first incarnation looked at consumer sentiment. Then marketing analytics. Then, around 2016, the team noticed something. The same machinery that could predict box-office returns from chatter about a movie could also predict whether a corporate counterparty was, in technical terms, a problem. They pivoted into financial crime. It is, as pivots go, not the most obvious one. It also turned out to be the right one.

By 2019, the partnerships started - Oracle, Acuris. By 2020, Snowflake and OpenCorporates. In 2021, the U.S. Department of Defense came calling. By early 2025, Quantifind had pulled together a $22 million round and launched a new product line aimed squarely at payments risk. The thesis held.

Graphyte: a knowledge graph with strong opinions.

Quantifind's platform is called Graphyte. Calling it "a screening tool" is technically correct and meaningfully wrong, the way calling a telescope "a tube" is. Underneath, Graphyte does four things that legacy systems do badly.

It resolves entities. When two news articles mention "J. Smith" and "Jonathan A. Smith Jr.," Graphyte decides whether they are the same person. The company claims roughly 90% accuracy on this, which sounds modest until you realize the legacy benchmark is closer to a coin flip.

It reads in many languages. Adverse media does not politely publish itself in English. Graphyte parses foreign-language content, which matters when the shell company you are trying to identify was incorporated somewhere with a different alphabet.

It builds a graph. Entities connect to other entities through directorships, addresses, transactions, news co-mentions. The graph is the point. A single suspicious payment is a data point. A suspicious payment from a counterparty whose director also sits on the board of a sanctioned shell - that is a story.

It explains itself. Regulators want to know why a model flagged what it flagged. Graphyte produces evidence trails. Black boxes do not pass audits.

The numbers are unromantic. The numbers are the point.

Quantifind sells to skeptics. Compliance officers do not buy enthusiasm; they buy reduced workload and cleaner audits. So the company has spent years collecting the kind of measurable claims that survive a procurement review.

Make illicit money visible.

It is worth saying plainly what Quantifind's customers do with this technology. A digital bank uses it so that a victim of a romance scam does not unknowingly become a money mule. A federal agency uses it to map a sanctions-evasion network across three jurisdictions. An anti-trafficking nonprofit named Polaris uses it to follow money that, in the wrong hands, follows people.

This is a long way from a movie-sentiment model in 2009. It is, however, the same underlying conviction: that the public record contains far more truth than any institution is currently extracting. The mission Quantifind keeps repeating - automate the discovery of financial risk by reading the world's unstructured data - sounds dry. The implication is not.

The next regulator is going to ask harder questions.

Two things are happening at once. AI is making it easier for criminals to construct convincing identities, generate plausible documentation, and move money in ways that look, to a 2015-era model, perfectly fine. And regulators - in the U.S., the U.K., the E.U., and increasingly elsewhere - are demanding more from financial institutions, faster, with more explainability.

That collision is Quantifind's market. Every bank in the world is going to need better tooling, and most of them are going to need it before they want to admit it. Whether Quantifind specifically wins the next decade depends on the usual mix of execution, distribution, and a few decisions about platforms and pricing. The thesis, though, is durable. Risk lives in unstructured data. Someone has to read it.

It might as well be the physicists.

The queue is still 1,400 alerts. But not for long.

That same compliance analyst, the one with the impossible Monday queue, logs in this week using Graphyte. The system has already ranked the alerts. The top twenty come with evidence trails - news coverage, corporate filings, network connections, foreign-language sources. The bottom 1,200 have explanations for why they were down-ranked. She finishes by Wednesday. The two she escalates turn out to matter.

This is not a story about replacing her. It is a story about giving her better questions to ask. That is what Quantifind, in the end, sells. Better questions, asked faster, at the scale a modern bank actually operates at.

The wordmark on the door at 444 High Street remains a particular shade of yellow. The two physicists are still there. The thesis still holds.

The official record.