Human + AI · Experience research
Understand how people experience your product.
AffectEQ reveals the probable human experience behind every screen — attention, effort, and a corroborating affect signal, each validity-gated and kept entirely private.
Analytics counts the clicks. We measure the experience.
How we measure
Four lanes of evidence. One clear picture.
Deterministic, AI, and human. Three independent kinds of evidence, kept separate and never blended into one fake number.
Deterministic audit
99 always-on expert usability checks (121-rule catalog). The same verdict every run, no AI, no internet.
AI visual review
A local vision model critiques the screen like a sharp designer, then verifies its own findings.
Live human affect
Facial expression, gaze, and a corroborating affect read from the face — validity-gated, on-device, never the cloud.
Cognitive load
Where your interface makes people work too hard. A per-person effort index.
In your pipeline
Code quality. Security. Performance. Now the experience.
Functional tests answer one question: does the machine's experience work? Nothing in your pipeline asks the other question — is it usable for the human on the other side? That's the gap AffectEQ fills: an experience-quality gate that scores every commit's built interface, locally, before a single user ever sees it.
Prove the machine's experience
Everything passes. The build is green. And a user still can't tell which field failed, or whether that grey text was a button.
Prove the human's experience
An expert verdict on the built interface — the kind a functional tool structurally cannot produce, delivered at the speed of a test.
Every changed page
On each commit or preview deploy, the engine renders the pages your change touched and reads what actually shipped — the built interface, not the mockup.
Expert evaluation, locally
A task-specialized model reads the screens the way a person would, cross-checked by a deterministic heuristics engine — so the trustworthy signal is the one allowed to block.
A verdict in your PR
Screen, flow, score, reason — the same grammar as your code-quality and security checks. Advisory until a threshold earns the right to hard-fail.
Nothing leaves your infrastructure — provably.
The models, the evaluation, the corrections — all local. There is no call-home, no telemetry, no cloud path. Air-gap capable, by architecture, not by policy.
Zero marginal cost per evaluation — every commit.
Once it's on your runner, each evaluation is effectively free. So it runs on every commit, not once per quarterly research study.
Works where the cloud doesn't.
Regulated, offline, and sovereign environments are first-class. If your code can build there, AffectEQ can gate there.
How we're different
World-class research. Human insight and AI, as one.
The depth of human understanding with the reach of AI. And not a single frame ever leaves your device.
Private by design
Faces, screens, and voices never leave the room. Local-first is how the system is built.
Rigorous, not guesswork
Every finding is grounded and reproducible. Re-run it yourself and reach the same result.
Honest about certainty
We tell you how sure we are, signal by signal. Never a guess dressed up as a fact.
Where we stand
A pro-human AI company.
We build AI to understand people, not to replace them. A real person is always in the loop, and every insight starts with their genuine reaction. Our AI only organizes what we measured. It never generates a feeling or invents a finding.
We consolidate the evidence, and we tell you how sure we are.
A new category has to earn its trust.
Nobody has proven that a model's judgment of usability is reliable enough to gate a release — so we don't get to claim it. We get to demonstrate it. Right now we're measuring our engine against panels of human experts on a frozen benchmark, and the first thing we'll publish is that evidence — agreement scores, failure cases, and all. Nothing customer-facing ships before the proof does. That's the honest deal, and it's the whole point of a gate you can trust.
Proof, not promises
Every finding is reproducible on your own machine. The same result, every time.
See how we prove it →