“I was like, ‘Guys, I’m tracking my stolen iPad. I know exactly where it is. Go get it,’” Delépine said. But they couldn’t because the police needed a detective to obtain a warrant, and they were all busy investigating the hundreds of other open cases.
Frustrated, he decided to build an AI tool that might change those odds. Now his software is vying to transform local police departments.
His San Francisco-based company — Longeye — was founded in May, and is now negotiating 20 contracts nationwide — including a few in San Diego — to help departments close cases.
With so much data and so few investigators, detectives don’t have time to comb through massive amounts of evidence, that’s why so many cases go unsolved.
In San Diego, 83% of sexual assaults, 33% of kidnappings, and 15% of murders went unsolved last year, according to the San Diego Police Department’s Annual Crime Statistics Report.
While detectives can’t sift through all the data, Delépine says, artificial intelligence can.
The Longeye platform ingests large volumes of digital evidence — cell phone data, jail calls, emails, PDFs, handwritten notes, images, spreadsheets and GPS data — and turns it into a searchable system for a single case.
Instead of spending days or months reviewing and organizing files manually, investigators can quickly surface the most relevant information.
While local agencies would not share details about how the software was deployed, Delépine cited a few Northern California agencies that solve cases using Longeye AI.
In a San Mateo County case, the software analyzed 537 jail phone calls in an open homicide case and flagged a single exchange.
“One person asks something like, ‘Just tell me where you shot her, was it in the head?’” Delépine said. “And the other responds, ‘Yep, somewhere up in there. But I really can’t talk about the case.’”
Hidden within hours of audio, just a few seconds were enough to qualify as a homicide confession.
In another case in San Mateo County, two robberies occurred about an hours drive from each other, a few days apart. Police believed they were connected and subpoenaed cell tower data, pulling in roughly 300,000 records of activity — everything from calls and texts to app pings.
Using Longeye’s software, investigators identified the phones that appeared in both locations at the same time.
Seven devices matched. Four were linked to telemarketing companies. Of the remaining three, all communicated exclusively with one another.
This data helps to build a pretty strong case, says Delépine.
Longeye is one of several AI startups offering technology to streamline surveillance and policing operations.
One of the most notable intersections of law enforcement and AI is Palantir, a $350 billion company. The tech behemoth works with the Department of Defense, Immigration and Customs Enforcement and the FBI, among other agencies, to collect large amounts of data, find patterns and prioritize leads.
Unlike Palantir, Longeye’s technology does not predict potential targets or criminal activity, but it does deliver case summaries and point detectives toward potentially consequential evidence.
“Longeye doesn’t replace their judgment. It just gets them to the right evidence dramatically faster,” said Denton Carlson, former police chief of the San Ramon Police Department in Contra Costa County.
After retiring, Carlson joined Longeye as a public safety ambassador, convinced the software could transform how departments handle overwhelming caseloads.
But critics warn against the embrace of AI policing technology.
“It’s a little scary to think about a government entity that has that much information about you,” said Amy Kimpel, associate professor of law and executive director of the Innocence and Justice Clinic at California Western School of Law in San Diego.
The only standing privacy protection in the age of information is the logistical implausibility of going through it all. “With this technology, that one protection gets eradicated,” she said.
The shortcuts, Kimpel warns, aren’t without consequence. “If you put in a prompt with thousands upon thousands of pages of data about someone, it’s pretty easy to develop something incriminating.”
In one case, Longeye software incorrectly flagged a phone call as proof of guilt. A suspect’s priest had read a biblical passage in Spanish about confessing sins, and Longeye pointed to that excerpt as a confession.
“AI will make mistakes, and we’re honest about that. The safeguard is our verification-first design,” Delépine said. “They listened to the 15 seconds. They were like, ‘OK, no, this is the Bible.’ And they moved on.”
But Kimpel wonders how many investigators would not move on if the error were less obvious. “We know investigators are putting in prompts looking for something incriminating, but what about information that is mitigating about this particular person?”
She doesn’t think the software is going to solve the problems that are inherent with policing in our society. “I’m not saying we shouldn’t use AI. I use AI. But with every new policing technology, there’s promise and peril.”
Delépine agrees, to a point. “If you believe the whole justice system is biased, that’s just a problem we’re not solving. The warrants are signed off by the courts, and this is where we go to eliminate bias.”
The worst cases of wrongful conviction tend to happen under pressure, Delépine explained, when officers are overworked, sleep-deprived and rushed. “All these mental shortcuts are our biases,” he said. “Fighting biases is really just the act of slowing down and not taking those shortcuts.”
And Delépine says his technology will allow detectives more time to conduct thoughtful investigations.
The El Cajon Police Department, which has a 30-person investigative unit, is currently piloting the software, though results have been limited. The department has deployed the AI on one case, but competing priorities have slowed progress. It’s “not any fault in the product,” said Lt. Tim Thorton of the El Cajon Police Department.
With the trial period nearing its end, El Cajon has yet to commit to a full deployment of Longeye AI.
The San Diego County Sheriff’s Office has been in a similar holding pattern. It’s unclear whether they are still actively using the software, but Lt. David Collins, the office’s media relations director, was receptive.
“It was one of the more impressive solutions we’ve seen recently,” he said. Like El Cajon, though, the decision ultimately comes down to money. “We are assessing where this solution falls within those parameters,” Collins said.
Agencies pay an adjusted rate depending on how many officers they have. A small agency of 20 officers would pay $5,000 per year, for example.
For now, Delépine is hitting the road and presenting Longeye to district attorneys and detectives across the country.
The company has raised $5 million to date, most of it from a venture capital fund owned by billionaires Marc Andreessen and Ben Horowitz, with a larger key first funding round on the horizon.
© 2026 The San Diego Union-Tribune. Distributed by Tribune Content Agency, LLC.
