Genesis Assistive Biometrics One core, many scenarios
Support
Genesis Lab Foundation

One sense.
Many scenarios.

A browser already has eyes and ears. We turn them into a single on-device sense — face, hands, posture and voice, read frame by frame and kept on the machine.

Everything below is that same sense, pointed at a different person's need. Nothing is uploaded, nothing is stored unless you deliberately save it, and every reading arrives as a suggestion with its evidence — for an adult to confirm.

See the scenarios Project overview
SENSING CORE Live, on this device. Points only — the camera image never leaves your browser.
What the core reads

Four signals, one pipeline

The same capture loop feeds every scenario. Change the question, not the machinery.

Face Dense landmarks, blendshapes, head pose, eye direction, mouth shape.
Hands 21 points per hand, handedness, openness, pinch, fingertips.
Body Pose keypoints head to feet, reach, turning away, leaving frame.
Voice Loudness, peak, pitch estimate, spectral centroid, voice activity.
The whole picture

One core, eight branches

Signals enter on the left, the core turns them into landmarks and features, and each branch asks its own question of the same stream. Follow an edge to jump to its block.

Face Hands Body Voice SENSING CORE on-device · nothing uploaded Nonverbal children Emotion and gaze Nonverbal audio Gesture layer Fingerspelling Mouth analysis Per-child calibration Wearable context
Built from blocks

The mission, and the seven layers under it

One block per scenario: who it is for, what it does, and the demo you can open right now. Where a demo runs on simulated data instead of live capture, the block says so.

02 Live camera

Emotion and gaze

For the caregiver watching alongside

A live read-out of apparent emotional state and where attention is pointing. Emotion and gaze only — no age, no gender, no identity signature.

Reads
Face landmarks, blendshapes, head pose, iris direction.
Gives
An affect estimate and a gaze vector, updated frame by frame.

A facial expression is not internal truth. Treat it as a prompt to check, not a fact.

Open demo
03 Live microphone

Nonverbal audio

For a child who vocalises but does not speak

Frame-level audio features feed a transparent heuristic that sorts sound into five acoustic categories: silence, distress, content, babble, vocalizing.

Reads
Loudness, peak, zero-crossing rate, spectral centroid, pitch, voice activity.
Gives
Live scores per category, with the feature values on screen.

Acoustic categories only. It does not detect emotion, intent or any clinical state.

Open demo
04 Live camera

Gesture layer

For reading the body when there are no words

Hand and pose landmarks turned into named gestures by pure geometry: pointing, hand-to-mouth, self-touch, waving, grasping — each with a confidence bar.

Reads
Hand landmarks and body pose, tracked over a short time window.
Gives
An active gesture plus per-gesture confidence.
Open demo
05 Live camera

Fingerspelling

For learners and for hands that sign before they speak

The same hand tracking, asked a different question: which letter is this shape? Live practice feedback with per-letter scores, entirely on the device.

Reads
21 landmarks per hand, finger curl and spread, handedness.
Gives
A detected letter, a confidence meter and per-letter scores.

Twelve static letters so far (A B C D E I L O U V W Y). Motion letters are not covered yet.

Open demo
06 Live camera

Mouth analysis

For speech therapy practice and early sound work

A viseme-lite mouth shape derived from blendshapes and lip geometry — a step beyond the plain mouth-open heuristic the scanner starts with.

Reads
Lip contour, jaw opening, mouth-related blendshapes.
Gives
A live mouth-shape class with the geometry behind it.

No tongue detection: no reliable on-device tongue model exists yet, so we do not claim one.

Open demo
07 Simulated data

Per-child calibration

For the child whose neutral is not everyone's neutral

Every child's resting face and voice sit at their own baseline. This layer captures a short baseline per child and reports how far a live reading has drifted from it.

Reads
Derived feature numbers only — no images, no audio kept.
Gives
A stored profile and a live z-score drift report.

The demo runs on a synthetic signal generator, not on live capture. The maths is real; the child is not.

Open demo
08 Simulated data

Wearable context

For the hours the camera never sees

Optional watch or band signals as coarse context around a session — consented per scope, and only ever context, never a verdict.

Reads
Heart rate, HRV, movement and rest, each behind its own opt-in.
Gives
A context state that sits beside the camera reading.

Simulated stream only. No hardware is accessed, and no signal here is a medical interpretation.

Open demo
Genesis Lab Foundation

Help build tools that help.

This is charitable work, built in the open and given away. Support funds the research, the validation with clinicians and families, and the layers still missing. Take your time — read first, and give only if it makes sense to you.