A detection score is an opinion.
A layered case is evidence.
Viskopic replaces the single opaque AI-detection score with three defensible metrics and a live viva — so compliance officers get an argument they can actually stand behind, not a number they have to take on faith.
A score is not a case.
Every AI-detection tool ends the same way: a single number, and no way to explain it. A 78% score isn't evidence, it's an opinion the compliance officer can't defend in a hearing, an appeal, or a tribunal. They can't say which sentences moved the number, why the model thinks so, or how confident they should be acting on it.
The people asking a university to prove misconduct don't want a probability. They want a case: something built from evidence a non-technical panel can read, question, and stand behind when a student pushes back.
"The detector said 82%. Legal asked what that meant. We didn't have an answer."
— composite of pilot conversations with university compliance staff
Three scores, not one.
We split the single black-box number into a layered read of the writing, across two domains that feed two different questions later in the process.
Consistency
Is the writing style of this author consistent?
Measured against that student's own submission history, not a generic population of writers.
Genericity
Is this author's style considered generic?
How closely the voice resembles the flattened, average style that generation models tend to produce.
Authenticity
Are the claims, reasoning, and evidence original?
Whether the argument in this specific piece, on its specific topic, holds up as this author's own reasoning.
Three numbers a compliance officer can point to individually, cite separately, and defend on their own terms.
When the scores flag, a conversation decides.
Hardcoded thresholds catch the obvious red flags. Everything past that line doesn't get a rerun of the same algorithm, it gets a live interrogation.
Threshold flag
A consistency or genericity score crosses a hardcoded line, an obvious red flag worth a closer look.
Style Viva
An AI probing system interrogates the student against their own past work in a similar field: "You wrote X in essay Y — why is it different now, in Z?"
Reasoning Viva
If style flags further, the AI generates the reasoning questions and a supervisor runs the session live, one-to-one, evaluating the student's answers in real time.
Watch a probe run.
Two transcripts, scripted from how each viva actually runs. Pick one to play it back.
The framework becomes the record.
Every case a university runs through Viskopic adds to a framework their own compliance officers actually understand, one built from that university's scores and viva outcomes, and cited in hearings and appeals as it accumulates.
That's the part a competitor can't hand over on day one. It takes a compliance office months to learn how to read and interpret a new framework in the first place, and every case decided on ours becomes a precedent their own staff already trust.
Rebuild the framework
Months of relearning what each score means and how it's been interpreted case by case.
Retrain compliance staff
Every officer who has learned to read our scores starts over on someone else's.
Risk open cases
Switching mid-stream can undercut the basis for cases still open or under appeal.
Where this goes from here
Metric engine & pilot
Ship the consistency, genericity and authenticity scoring engine and run it with a small number of pilot institutions against real submissions.
Viva infrastructure
Build the Style Viva and Reasoning Viva flow, threshold triggers, the AI probing dialogue, and the live hand-off to a human supervisor.
Compliance framework API
Open an API that lets an institution's own framework, its scores, thresholds and precedent, be queried, cited, and carried forward.
Who's building this
AI Engineering, Lloyds Banking Group. BSc Data Science, University of Bristol.
MSc AI & BSc Data Science, University of Bristol.
BSc Mathematics & Computer Science, University of Bristol.
Let's talk
We're pre-seed and talking to early pilot partners and investors who think about this problem the way we do.