Asymmetric verification, explained.
A resume is a set of claims. Provicio checks whether the person making them can talk about them unprompted. Here is exactly what happens, what comes out the other end, and where the method stops working.
Resume becomes claims
You upload a candidate's resume. Provicio parses it into discrete, checkable statements: roles held, systems built, scale handled, outcomes claimed, technologies used. Vague lines (“team player”) are ignored; specific ones (“reduced p99 latency 40%”) become claims.
Candidate talks, resume stays closed
The candidate gets an email in your name, reads a plain-language consent screen, and answers a short set of open-ended spoken questions about their work: what they built, what broke, what they'd do differently, what they were personally responsible for. The questions never quote or reference the resume. Eight to ten minutes, on their schedule, no coding.
Transcript compared to claims
Audio is transcribed and deleted. The candidate reviews and can correct their transcript before submitting it. Then each claim is checked against what was actually said and marked supported, unsupported, or unverified, with a sentence explaining why. An overall high, medium, or low confidence level summarizes the picture.
Why asking about the work, not the resume, is the whole trick
If you ask someone “tell me about the Kubernetes migration you led,” you've handed them the answer's outline. Anyone who has read their own (or their resume writer's) document can improvise from there. If instead you ask “walk me through the most complicated infrastructure change you were personally responsible for, and what went wrong,” the person who did it talks about the dual-write window, the DNS cutover that failed on a Friday, and the teammate who caught it. The person who didn't do it talks in generalities, changes the subject, or describes a different, smaller thing.
Provicio scores the presence of specific, consistent detail that lines up with the claim. It does not score fluency, confidence, accent, vocabulary, or how polished the answer sounds. A halting, heavily accented answer with the right details is a supported claim. A smooth, articulate answer with none of them is not.
What you get back
- Per-claim verdicts with reasoning. Not a score you have to trust; a list you can check in two minutes.
- An overall confidence level. High, medium, or low. It's a sort key for your attention, not a decision.
- The transcript. As reviewed by the candidate. If you want to read what they actually said, it's there.
- A record. Who ran the check, when, what was flagged, and what your team decided, exportable and retained for four years.
What it catches
- AI-inflated histories. A real career with invented scope, seniority, or metrics layered on top. The most common case.
- Resume-mill and template resumes. Documents optimized for a job description that the applicant didn't write and can't speak to.
- Borrowed experience. A teammate's project, a former employer's architecture, or a shared team resume presented as personal work.
- Applicants who don't know the work they claim. Including a large share of organized-fraud profiles, which lean on stolen or fabricated histories and rely on nobody asking follow-up questions.
What it doesn't catch
We'd rather you hear this from us than discover it later.
- A fluent, prepared liar. Someone who has memorized a fabricated history in enough depth to improvise around it for eight minutes will pass. Provicio raises the cost of faking a resume from “ask an LLM” to “actually learn the work,” which eliminates most of the volume, but not all of it.
- A proxy, on its own. The person who records the check may not be the person who shows up later. Your interviewers still need to confirm they're talking to the same person. Using the recorded session as a baseline to compare against the interviewee is the extension we're building toward.
- Skill level. A candidate can be entirely honest about their resume and still not be the engineer you need. That's what interviews and skills assessments are for. Provicio's job is to make sure those are spent on real people.
- Identity documents, employment history databases, or background checks. Provicio verifies that the candidate can account for their claimed experience. It does not verify who they legally are or call their former employers. Pair it with your existing background-check process; don't replace it.
Where it fits
Wherever you decide. Provicio is one asynchronous step triggered by a single link, so it doesn't require you to redesign anything. Teams can run it on every candidate who clears a first conversation, reserve it for resumes that look too good, or use it as a final check before a submittal or an onsite. The economics are the same anywhere: every unsupported claim you catch is time you didn't spend downstream. Volume varies with how you use it, which is why we price flat rather than per candidate.
What it's not
Provicio is not a lie detector, does not analyze voice stress or biometrics, does not keep audio, does not produce a hire/no-hire recommendation, and never auto-advances or auto-rejects a candidate. It is a structured way to ask “can you tell me about this?” at scale, with the answers organized so a human can review them fast. Everything about how data is handled is on the trust page.
Want to see it on a real resume?
Bring a resume you already have doubts about to the call. We'll walk through what Provicio would ask and what it would look for.
Talking to the engineer who built it, not a sales team. If it's not a fit, you'll hear that on the call.