The interrogation room has not changed much since the Cold War. A table, two chairs, a recorder, and a human trying to remember what to ask next while reading a face. Prometheus Intelligence, founded in 2025 in Newnan, Georgia, wants to put a second brain in that room, one that plans the next question while the first brain watches the eyes.
The company calls its product Prometheus Torch and describes it as the first AI-enabled interviewing system built on dynamic questioning tradecraft rather than simple transcription [prointel.net]. The distinction matters. Transcription is a stenographer. Tradecraft is a colleague.
The bet on tradecraft over transcription
Most of the AI tooling sold into law enforcement and intelligence so far has been recognition: transcribe the audio, translate the language, flag the keyword, search the database. Useful, but passive. Prometheus is selling something more ambitious, a multi-agent architecture that plans, guides, and analyzes interviews in real time [prointel.net].
In practice, that means an agent suggesting the next question based on what the subject just said, another scoring sentiment and consistency, and a third handling translation when the subject is not speaking English. The pitch is that a junior investigator with Torch in their ear performs closer to a seasoned one. Whether that pitch survives contact with an actual field office is the open question, and it is the question the next twelve months will answer.
Why the founders are the story
Co-founders Nate Boaz and Randy Meyer met at the U.S. Marine Corps Counterintelligence and Human Intelligence school in 1999 and served together through Operations Enduring Freedom and two Iraq War campaigns [prointel.net]. Their team's work, per the company, contributed to the rescue of seven American POWs and the capture or elimination of high-value targets [prointel.net]. The Newnan Times-Herald confirms both were part of the early invasion forces in Iraq [times-herald.com, 2026].
That biography is not decoration. In a category where the buyer is a defense or intelligence program officer, founder credibility on the tradecraft itself is a meaningful wedge, often more meaningful than the model architecture under the hood.
| Founder | Role | Background |
|---|---|---|
| Nate Boaz | Co-founder, CEO | USMC CI/HUMINT, 1999; OEF and Iraq War [theorg.com] |
| Randy Meyer | Co-founder, Chief Client Officer | USMC CI/HUMINT, 1999; OEF and Iraq War [theorg.com] |
Where the headwinds live
The company is early, and the cited evidence reflects it: no disclosed funding, no named customers, no announced partnerships, and 629 followers on LinkedIn at the time of capture [LinkedIn]. The govcon directory lists the entity with a UEI and CAGE code, the basic plumbing for federal contracting, but a CAGE code is a starting line, not a contract [GovCon in a Box]. The real risks cluster in three places.
- Procurement cycle. Defense and intelligence buyers move on timelines measured in fiscal years, not quarters. A 2025-vintage company without disclosed funding has to survive the gap between pilot and program of record.
- Evidentiary standards. An AI that suggests questions in a law enforcement interview will eventually be cross-examined in a courtroom. The bar for explainability is different from the bar in a commercial workflow tool.
- Incumbent reach. The existing language and analytics vendors already inside the Pentagon and the Bureau have the contract vehicles, the clearances, and the integrator relationships. Torch has to be enough better at one specific thing to displace them.
The unit math, roughly
Work the back of the envelope. A federal linguist contract runs in the low six figures per analyst per year fully loaded, and a single agency office might field dozens of interviewers across counterintelligence, immigration enforcement, and federal law enforcement. If Torch can credibly take one hour off a four-hour interview workflow and shave the post-interview report from a day to an afternoon, you are looking at something like 600 to 800 reclaimed hours per analyst per year. At a blended internal rate of $120 an hour, that is roughly $75,000 to $100,000 of recovered capacity per seat, against a software price that almost certainly lands between $15,000 and $40,000 per seat (estimated). The ROI math is the easy part. The hard part is the accreditation paperwork.
The incumbent Torch has to beat is not another startup. It is SRC's and SAIC's language and analytics stack, the quiet apparatus that has handled translation, transcription, and tradecraft tooling for the intelligence community for two decades. Those vendors do not have a tradecraft-native interviewing agent. Two Marines in Newnan, Georgia think that is the gap worth walking through.
Sources
- [prointel.net] Prometheus Intelligence home and Our Story | https://www.prointel.net
- [theorg.com] Prometheus Intelligence leadership | https://theorg.com
- [LinkedIn] Prometheus Intelligence company page | https://www.linkedin.com/company/prometheusintelligence
- [GovCon in a Box] Prometheus Intelligence Inc listing | https://www.govconinabox.com/explore/smb-federal-contractors/prometheus-intelligence-inc-PW8DT7RK5261
- [times-herald.com, 2026] Coverage of Boaz and Meyer | https://times-herald.com