Meet the small group of engineers helping the public sift through the Epstein files 

After officials released millions of pages of documents related to the late sex offender Jeffrey Epstein, revelations in his emails and other files have led to the resignations of multiple corporate executives, new investigations into abuses by Epstein and potential accomplices, and even the arrest of the United Kingdom’s former Prince Andrew.

For those looking to research Epstein’s vast correspondence and web of connections across industry, government, and academia, some of the most effective tools have been built not by federal investigators or big-name news organizations but by a scrappy team of volunteer developers.

Starting with a website called Jmail, which made Epstein’s publicly released emails searchable through an interface cheekily copied from Gmail, they have since built a set of web apps modeled after familiar sites like Google Drive, Wikipedia, Amazon, and YouTube. The goal: to turn messy PDFs and other files released in bulk by federal officials into something members of the public—including journalists—can more easily search and understand.

Key to the project’s speedy success is the technical talent of the team of around 15 named core contributors. But equally vital, they say, is the current wave of AI tools that helped them rapidly generate code and process huge troves of data.

“So not only do we have an app that we were able to make very quickly, we have data that can populate that app with real content,” says Luke Igel, among the project’s initial creators. “Both those things had to come together; both of those were not possible a few years ago.”

Igel, an MIT grad who is cofounder and CEO of video software company Kino, says the inspiration for the project came after he and a friend were discussing an initial tranche of Epstein-related documents released by members of Congress in November. They were struck by the extent of Epstein’s ties to political figures across party lines and around the world but questioned whether the public would be able to fully understand the story as the data was initially presented.

Igel then reached out to Riley Walz, a developer and entrepreneur known for creative internet projects (including a recent parody of Apple’s “Find My” interface that tracked San Francisco parking enforcement officers) about collecting the emails in a Gmail-style interface.

Thanks to AI development tools like Cursor and Anthropic’s Claude models, the pair was able to put together the first version of Jmail in just a few hours, Igel says. “We cloned Gmail, except you’re logged in as Epstein and can see his emails,” Walz announced in a viral X post in November.

When the Department of Justice released an additional trove of files in December, spurred by the Epstein Files Transparency Act passed by Congress the previous month, a group of about 10 collaborators gathered at Igel’s San Francisco home and via video conference to build the next iteration of the software.

The team also had help from a company called Reducto—a maker of software that turns messy PDFs and other complex documents into structured data—to parse the newly released files, which had become too complex for general-purpose AI tools to decipher reliably.

“A lot of these PDFs are scans of printouts or handwriting,” says Adel Wu, who works on growth at Reducto. “It was actually very messy.” 

The company—which is located in the same building as Kino—had already been considering doing something with the Epstein files and quickly decided to support the Jmail effort after hearing about it, says founding engineer Omar Alhait, noting, “We very quickly went through all of the documents and parsed out all relevant email information from them.”

Reducto’s software helped accurately render redactions within the documents and even let the team extract complex information like Epstein’s flight data, which was made available in a Google Flights-style interface called JFlights. Again, AI—including Anthropic’s then-new Claude Opus 4.5 model—helped the Jmail team rapidly develop new features and apps and merge thousands of code updates in a short time.

“So much of what I thought was core to software engineering is actually something that this model can help you with and help you blast through very quickly,” Igel says.

The team’s investment in infrastructure let them quickly import, process, and share additional documents released just before Christmas, though the project drew even more attention after a massive DOJ release of millions of Epstein-related files on January 30. Handling that release required not only processing the new documents—Alhait says it took Reducto about three days to crunch through the data—but also beefing up the project’s infrastructure to handle an influx of traffic as public interest in the files continued to grow.

“Tons of people came to the house again, and this time we really just had to make it scale,” Igel says. “Everything broke. Tons of scaling issues we thought we had solved, with database outages and caching failing, came through again.” 

With the help of AI tools, the team stabilized the site, which has now served more than 500 million page requests to more than 50 million unique visitors. The project has also expanded beyond Jmail and JFlights to include an AI guide to the files called Jemini, a video repository called JeffTube, a file repository known as JDrive, and even a searchable log of Epstein’s Amazon orders called Jamazon.

The team works to ensure information in the files is properly redacted to protect sensitive details, taking care to update the site’s available materials to reflect any new redactions by federal officials. “It’s very, very important to us to be as responsible as possible when surfacing information to the public,” says Melissa Du, an AI research engineer who works on the project. “We obviously don’t want to be over-redacting, but also the privacy of the victims is of utmost importance.” 

Du, another MIT grad, says she became “morbidly fascinated” by the first set of files released on Jmail, including documents referencing MIT-linked academics such as former Media Lab director Joi Ito and professor emeritus Noam Chomsky. She has since worked on aspects of the project such as JDrive for data management and the Wikipedia-style Jwiki, which was first populated with write-ups of key Epstein-linked figures generated by AI and then carefully vetted before publication.

Perhaps most striking about the project is that a small group of developers was able to do what major media organizations had done in organizing previous viral data repositories, like former intelligence contractor Edward Snowden’s revelations about government surveillance or the offshore finance leaks known as the Panama Papers.

The team has received about $32,000 in donations to cover various costs, along with donated technical services from Reducto, Kino, and cloud provider Vercel. But the core work has been carried out by developers with their own day jobs and startups.

Though at times Igel wondered whether the project would be effectively scooped by big news organizations building their own Epstein data explorers, data from the Jmail project has actually been cited by news outlets including The Economist. The team has also been in touch with congressional staffers about passing on crowdsourced requests for release of potentially excessively redacted files.

And additional features are being considered, including a Google Calendar-style interface to explore calendar data in the repository, says Igel, who notes that the underlying code from the project will also likely be released as open source in the future.  

Already the project stands as an example of what’s possible for a talented team equipped with the latest in AI development and data processing tools. “We’ve really relied on the new AI models,” Du says. “And we’ve also just had a very high level of trust across the team.” 

View the full article