VERITROOPER makes AI honest on your own data — and proves it. We measure how accurately any AI answers from your documents, show your team exactly where and why it’s wrong, and hand back audit-ready evidence — against a frontier model or an assistant you’ve already deployed.
A deep, point-in-time audit of how accurately a model answers from your documents — every wrong answer, exactly why, and audit-ready evidence to back it.
A deep, point-in-time audit. Measure how accurately any AI answers on your data, find every wrong answer and exactly why, and get the achievable ceiling — the gap is the accuracy you’re leaving on the table. Runs as a self-contained app on your machine, against a frontier model or an assistant you’ve already deployed. Optional EU AI Act conformity evidence in one toggle.
Explore Scout →VERITROOPER is an accuracy engine for AI working over your own documents. The failure it exists to catch is hallucination — when a model confidently produces a wrong answer. Models don’t crash when they hallucinate; there’s no flag, no warning, no error code. They just sound certain about something that isn’t true. That’s what breaks naked AI deployment anywhere the answer actually matters.
VERITROOPER doesn’t replace your AI — it measures it, proves it, and tells you exactly what to fix. It generates a question set from your material with verified ground truth, scores the model against it, routes every failure to specialist diagnostics, and confirms contested verdicts with an independent verifier from a different vendor — the model under test never gets the final say on its own answers. It ships as Scout — a deep, point-in-time audit you run as a self-contained app on your own machine.
A result is only worth as much as the process behind it. Every step that produces one is built to be defensible — to your auditors, your buyers, and your own engineers: independent cross-vendor verification, scoring that rounds against us, built-in hallucination traps, every failure shown in full, tamper-evident sign-off. Integrity here isn’t a claim — it’s the mechanism.
Don’t take our word for it.
Just me. Brian B. Average guy. Veteran of the US Army and Air Force.
Before this April, VERITROOPER didn't exist — not as a name, not as a product, not as code. I was working on a D&D game to occupy my mind (disabled vet, had time). I taught myself Unity from zero, wired up an LLM as the dungeon master so the game could run scenes and adjudicate rulings. I had one nagging problem…
The AI couldn't catch me lying. I could feed it anything and it would confidently run with it. I could spend gold the character didn't have, cast spells that didn't exist, and perform actions the character couldn't actually perform. I wanted a DM AI that called my crap. It turned out no such thing existed. Trying to solve this problem led to VERITROOPER.
As it turns out, the DM not knowing, and therefore making answers up, was a far broader problem called hallucination. The DM, like any LLM-based narration today, would try to answer factually, and if it couldn't it would answer confidently, whether right, wrong, or just completely made up. That's AI hallucination in a sentence. I had no idea at the time.
As I zeroed in on my goal, I began to think it was possible. I was talking about it to a good friend of mine, who offhandedly said "I'd be really impressed if you could make it run RIFTS" (another gaming system). I said "I think I can." One thing led to another, and the realization that this system I had created worked on any data set, using any LLM began to set in.
D&D SRD was the first data set I ever ran when I built this — harder than it looks, and it bootstrapped the whole architecture. Any verification layer that can reliably handle rules-lawyering on game mechanics generalizes to tax code, military law, financial filings, and Air Force training material — and that's what the data sets in Scout’s results prove.
VERITROOPER is the unintended discovery of a bored man trying to make a D&D game. Inception to deployment-ready development happened in a short amount of time from a guy in his basement with a PC, a problem, and time on his hands.
For acquirers, pilot partners, and technical reviewers.
Live walkthroughs by request. Raw logs, full dataset, and patent package available on contact under NDA.
Pilot partners: a pilot runs VERITROOPER against your model and your data on a question set you approve, and returns the full audit — per-question failures, failure-recovery rate, and concrete fix recommendations. The public results, the sample run records, and the methodology need no NDA; raw logs, the full dataset, and the patent package are shared under NDA.
