He built Epstein’s ultimate database, and it consumed his life

Epstein Library On the Ministry of Justice website is a sample of the chaos. In early December, Keller was leafing through tens of thousands of pages of documents in the library, feeling “frustrated and in disbelief” at the chaos — files that could be hundreds of pages long, text that was sometimes blurry or sideways, a bank transfer without context, an email chain with half the names blacked out, a flight log containing only initials. “It’s confusing,” he says. “You’re reading parts of something enormous and trying to figure out what parts are important and how they connect to each other.”

One night, he spent about four hours trying to track down one person’s name across about 30 documents in the archive. “I just stopped and thought I was manually doing what a database could do in milliseconds,” he says. As a database infrastructure builder at a mid-sized company, he knew exactly what to do next. “I opened up the code editor and started building,” he says. “By 3 a.m., I had a basic research prototype running on a few hundred documents.”

Around that time, a site called Jmail.world It was making a lot of noise as a tool for people to view Epstein’s emails as if they were using the Gmail interface. Launched in mid-November and built by a group of tech-savvy volunteers, it has since evolved to include, among other things, his photos, flights and purchase history on Amazon, and is also displayed as if the reader were viewing Epstein’s private accounts. Keller used the tool and liked it. “Jmail was proof that the community can build better tools than what the government provides,” Lee said.

It also helped him hone his own project. “Instead of thinking about one category of documents, I started thinking about the network,” he says. “How do you link a person who appears in an email to a flight they were on, or to a bank transfer, or to a certificate they gave? This cross-referencing problem is what I wanted to solve.”

Then, on December 19, the Justice Department released the first large batch, adding hundreds of thousands of new documents to the existing archive. Immediately, Keller’s workload swelled to an all-time high. The prototype he built earlier in the month became the basis for processing all of this.

Most nights he worked until 3 or 4 a.m., sipping cold coffee while scrolling through a sea of ​​open tabs.

Because of his childhood, he says, “when the first documents started falling out, I couldn’t look away. I intuitively understood what was described in those files.” In the evening, he would come home from his day job, and once his family was all in bed, he would hide in his home office and spend hours browsing through downloaded PDF files.

Many documents were published as images, and he would run each page through layers of software to convert it into searchable text, and sometimes one system would fail to convert the text, running it through a second or a third. It then uses another system to extract important details such as names, organizations, dates, and locations. He was performing hash checking — a process that checks whether Justice Department files have been tampered with — and redaction analysis, looking for inconsistencies in how the government was withholding information. He kept track of all his work in an accurate, color-coded digital ledger. “It doesn’t download files,” he says. “It’s reconstructing the crime scene from two million pieces of evidence.”