An AutismAI Lab application
Continuous, clinically-aligned learning data for children with autism.
Grovalin is a tablet-based learning app for children with autism spectrum disorder. Every game, story, and activity is gated to the child's current PECS phase, while every practice attempt is tagged in real time against a VB-MAPP skill area and level. The result: a continuous, longitudinal, per-skill picture of what the child can do, refreshed at the end of every session.
Six PECS phases drive what every game shows. A child never sees content above or below where they actually are.
Five skill areas, multiple levels, one definition of mastery. The dashboard updates as the child plays.
The child practises on real objects from their own home, plus AI-generated colour and structural variations of each one.
Why this matters
Attacking generalisation at the data layer.
Children with ASD often master a skill in one setting and lose it in another. A child who can label "chair" looking at a dining-room photo may freeze when shown an office chair, a kitchen stool, or the same chair from a different angle. This is a generalisation problem, and it is the single most documented barrier to durable language gains in ASD intervention.
Grovalin addresses generalisation directly. Every "chair" the child practises on is a real object from their own home (extracted from a caregiver-uploaded photo or video), plus AI-generated colour and structural variations of that chair. The child learns that "chair" is a category, not a single picture, by interacting with many shapes of the same word. The clinical question then becomes: how does the BCBA know it is working? That is what the platform is built to answer.
PECS-phase-gated, end to end
One number on the child's profile drives everything they see.
Grovalin organises its entire learning experience around the six phases of PECS (Picture Exchange Communication System). The child's phase is one field on their profile, and it controls every game, prompt, and difficulty level the app surfaces.
| Phase | What the child is learning | What Grovalin shows them |
|---|---|---|
| 3 | Picture Discrimination Pick the right picture out of several. | Words mode and Object Game ("Tap the chair"). Difficulty scales from 2 to 6 distractors. |
| 4 | Sentence Structure Combine "I want" with a noun, then an attribute plus a noun. | Sentence Builder Level 1 ("I want chair") and Level 2 ("I want the red chair"). |
| 5 | Responsive Requesting Answer "What do you want?", not just request spontaneously. | Same games as Phase 4, with prompts shifting toward responsive language. |
| 6 | Commenting Describe what something is or what colour it is. | Colors Game ("Tap the red one") across many objects. |
Phases 1 and 2 of PECS are about the physical act of exchanging a card with a communication partner. Those skills require a human partner in the room and cannot be taught by a tablet. Grovalin meets the child where the app can actually help: Phase 3 onward. Phases 1 and 2 remain visible in the phase selector but greyed out.
Grovalin recommends advancement only when three independent conditions meet simultaneously: quality (80%+ accuracy), volume (enough trials), and distribution (across multiple days, objects, and difficulty levels). The caregiver makes the final call. Grovalin recommends; the human approves.
VB-MAPP aligned at the data layer
PECS gates what the child sees. VB-MAPP organises what we learn about them.
Every practice attempt is tagged at write time with two pieces of metadata: a VB-MAPP skill area and a level. The tagging is automatic, in real time, with no caregiver input.
| Skill area | What it captures | Where in the app |
|---|---|---|
| Tact | Expressive labeling: child sees an object and says (or hears) its name. | Explore Mode: tap an object to hear its label spoken aloud. |
| Listener Responding | Receptive language: child hears a label and selects the matching picture. | Object Game and Colors Game ("Tap the X"). |
| Visual-Perceptual / MTS | Matching-to-sample, visual discrimination across distractors. | Object Game at lower difficulty levels (smaller arrays, easier distractors). |
| Intraverbal | Verbal behavior elicited by other verbal behavior, including sentence construction. | Sentence Builder, both Level 1 and Level 2. |
| DLS (Daily Living Skills) | Self-care, hygiene, mealtime, dressing, household routines. | LifeSkills module, with each task template carrying its VB-MAPP DLS tag. |
A cell (or an object) is mastered when the child reaches 80% accuracy or higher across the most recent 3 sessions in which it was practised. The same rule applies to object-level mastery and to VB-MAPP cell-level mastery. There is one definition of "mastered" across the entire platform, enforced by an automated test that fails the build if the two ever drift apart.
How a BCBA uses the data
Three places the clinical signal surfaces.
Two cards side by side: a PECS card showing current phase, accuracy, and the "Ready to advance" badge when criteria are met; and a VB-MAPP panel with per-skill-area, per-level cells, configurable time windows (7, 30, 90, 365 days). Caregivers routinely screenshot it for clinical meetings.
One click produces a single-page report sized for a clinical session: child first name, time window, generated timestamp, one section per skill area in canonical order, per-cell rows with accuracy and mastery labels. Filename convention {child}_vbmapp_{YYYY-MM-DD}.pdf, ready to drop into a chart.
At session-end, pure detectors write barrier records when patterns emerge: prompt dependence (right answers, but only after prompts), weak discrimination (errors clustering on a specific wrong label). Records carry severity, first-observed timestamp, and evidence count. Auto-clear after 14 days without re-trigger.
What measurement-across-time looks like
The part paper data sheets cannot match.
Daily progress rollup: total questions, correct answers, response times, new masteries. Atomic and idempotent; retries do not double-count.
Daily VB-MAPP snapshot at session-end. At-most-once per UTC day; subsequent sessions extend rather than overwrite.
Mastery status, accuracy curve, sessions-tested count, last-tested date, average response time for every object ever practised.
VB-MAPP panel renders cumulative cell-level performance over the chosen window. Trends within an area visible at a glance.
Home card shows lifetime VB-MAPP skills mastered. The headline number a caregiver or clinician sees on opening the app.
A BCBA returning to the data after a one-month gap can answer, without rerunning probes: which skill areas the child has practised, which cells have crossed mastery and when, which cells are progressing but not mastered, which barriers are active, and how this month compares to last. No extra data collection burden on the caregiver, no extra session burden on the child.
Why this is different
VB-MAPP aligned, never VB-MAPP assessment.
| Approach | Strength | Weakness Grovalin addresses |
|---|---|---|
| Paper VB-MAPP probes | Clinician-administered, formally validated. | Discrete events, not continuous; high caregiver burden between sessions. |
| Token / generic learning apps | Cheap, engaging, accessible. | No VB-MAPP alignment; no PECS gating; no longitudinal per-skill data. |
| In-clinic ABA programs | Direct trial-by-trial measurement. | Confined to clinic hours; child practices on clinic stimuli, not home objects. |
| Grovalin | Continuous, VB-MAPP-tagged, PECS-phase-gated practice on the child's own home objects, with clinician-readable reports at session-end. | Not a substitute for clinical assessment. Designed to inform the BCBA, not replace them. |
Data privacy and parental consent
Built for children, designed for caregivers.
The platform is designed around COPPA-style principles: verifiable parental consent before any data is collected, full caregiver control over the data once it exists, and clean audit trails for every consent action. The very first screen of onboarding is a consent step. A child profile cannot exist without a matching consent record.
One-click export of a complete machine-readable archive: profile, sessions and activities, uploaded videos and images, extracted objects and variations, stories generated, full consent history. Logged as data_exported.
One-click account deletion permanently removes every database record and every S3 file tied to the account. Returns a statistics receipt so the caregiver sees exactly what was removed.
Toggling consent off immediately blocks new sessions and stops activity logging, while preserving existing data so consent can be restored later without losing progress history.
Consent is checked at runtime, on the server, at every write boundary: session start, activity logging, barrier detection. A withdrawn caregiver cannot start a child session, even if the client tried.
- The child's data does not enter Grovalin until the caregiver explicitly consents at sign-up.
- The caregiver can export everything Grovalin has stored, in a readable format, at any time.
- The caregiver can delete everything Grovalin has stored, permanently, at any time.
- The caregiver can pause new data collection without losing the progress already gained, at any time.
What Grovalin does not claim
Three explicit non-claims.
Grovalin is PECS-phase-gated. The card exchange, the human partner, the protocol fidelity stay with the trained therapist or caregiver, not the app.
Grovalin is VB-MAPP aligned. Dashboard numbers are signals from practice data; they are not scored milestones and do not substitute for a licensed VB-MAPP administration.
Every advancement, barrier flag, and report exists to inform the human in the loop. The caregiver approves phase changes. The BCBA interprets barriers.
Roadmap
Deeper clinical signal, wider collaboration with BCBAs.
- →More barrier detectors: low motivation, repetitive responding, weak generalisation.
- →Phase 5 full instrumentation: prompt-type differentiation at the data layer.
- →Tact exposure metric, separate from Listener Responding accuracy.
- →Generalisation sub-metric across colour and structural variations.
- →Therapist invite codes for read-only BCBA access.
- →BCBA-tunable mastery thresholds and barrier severity bands per caseload.
- →Sticky and manual barrier management tools.
- →Sign-off on the difficulty-to-level mapping driving VB-MAPP tagging.
- →Pilot studies comparing Grovalin-derived signals against clinician probes.
- →Shared library across caregivers for processed objects and stories.
Try Grovalin, or partner with us.
Caregivers can sign up and onboard a child in minutes. BCBAs and clinical practices interested in evaluating Grovalin for a caseload can reach the team through the collaboration page.