JC SEN Portfolio
☰
🧪 Demonstration · synthetic data only — no real children. Switch role

About the programme

JC SEN Portfolio is one part of a five-year programme for neurodivergent primary pupils in Hong Kong. This page explains where the programme comes from, who does what in a school, what the platform does step by step — and what it deliberately does not do. Everything in the demo uses synthetic data only; no real children.

Draft content 8 item(s) marked "to be confirmed" by the programme office.

Background

In Hong Kong, children with special educational needs (SEN) learn in ordinary primary schools alongside their classmates. This is called inclusive education.

The Hong Kong Jockey Club Charities Trust initiated a five-year programme, *Fostering an Inclusive and Empowering Learning Environment for Neurodivergent Primary School Students*. *Chinese title of the programme: to be confirmed.*

The JC SEN AI-Powered Student Portfolio Platform — *JC SEN Portfolio* in the app — is one part of that programme. It follows the programme's five guiding ideas: neurodiversity is variation, not deficit; build on strengths; Universal Design for Learning; the social model of disability; and learning as a social, shared process. The platform is an instrument of these ideas, not their owner.

Objectives

The programme's aim is in its name: an inclusive and empowering learning environment for neurodivergent primary pupils. Within it, the platform is portfolio-first and strength-based. It is not a diagnostic tool. AI is a small helper on top of trustworthy records — never the product — and every core task works with the AI switched off. It is designed to WCAG 2.2 Level AA and to principles that suit neurodivergent users: predictable navigation, one main action per screen, calm visuals, no auto-play motion. Here is what it is built to do.

  • One structured profile per child — One long-term portfolio per child — evidence, observations, insights and reports in a single timeline. It is the record the next teacher inherits.
  • Growth tracked in many dimensions — Progress is followed across areas such as communication, social skills and self-management — against the child's own past, never against other children. No scores, no grades.
  • Data entry anywhere, anytime — A moment can be captured on a phone in seconds — one-handed, and with the AI off if the school prefers.
  • Earlier, clearer signals — Systematic observation records give teachers earlier and better-organised signals. The platform itself never labels a child.
  • Home-school collaboration — Families see teacher-chosen highlights and approved reports, in plain language in three languages — only with consent.
  • Plans that stay human-owned — Goals, interventions and review cycles are written and owned by people, with version history and approvals. The portfolio timeline is the evidence behind each review.

School-based support model

Hong Kong schools follow a Whole School Approach — supporting pupils with SEN is everyone's job. They also use a 3-Tier Support Model: good teaching for all (Tier 1), small-group help (Tier 2), and intensive individual support such as an individual education plan (Tier 3). The platform's record of evidence and goals helps people make these decisions. It never decides a tier itself. This is why there are seven role-scoped views. Below is who does what.

Teacher / SENCO
Teachers capture everyday moments and write dated observation records. They edit and approve every AI draft in their own words, and choose what to share. The SENCO — the teacher each Hong Kong primary school appoints to coordinate SEN support — uses the same login but sees more pupils; anything that needs a look is passed up to them. Individual education plans stay human-owned.
Parent / Caregiver
Parents and caregivers decide, through consent, who may open their child's folder — and can withdraw it. They see only what the teacher has chosen to share: highlights and approved reports, in plain language. Children take part with an adult beside them, in a read-only, strengths-only view. *The consent model in the demo is the delivery team's design; the agreed model is to be confirmed.*
School admin
The school admin sets up the roster, staff, classes and consent records — metadata only. This role has no route into any child's portfolio: content, media, insights and reports are out of reach by design.
Programme admin
The programme admin sees a live, de-identified dashboard across all schools. This role holds the levers that keep the privacy promises: which AI features are on or off, the small-cohort threshold, the languages offered, and the consent-policy wording. The platform also keeps help and training records, including completion of an AI-security workshop for SENCOs. *Which organisation will hold this role: to be confirmed.*
Professionals and partners
Educational psychologists and occupational therapists review sensitive outputs; their professional judgement has the final say. NGO coordinators and researchers see de-identified figures only — never names, never media.

How it works

The whole journey, end to end — the same evidence, seen the right way by each person. A simple picture helps: a student is a folder; an evidence item is a page filed into it; a report is a bound stack of chosen pages; the AI is an assistant who reads the folder and drafts new pages the teacher signs; consent decides who may open the folder.

  1. Capture
    The teacher logs an everyday moment — a note, a photo, or a short clip. Two flags decide whether the child may see it and whether it may be shared with the family. It can be tagged to a framework goal.
  2. Insight
    For a clip, the system derives gentle, non-diagnostic signals — for the teacher's eyes only. They are derived signals, not raw video: a prompt for the teacher's judgement, never a diagnosis. *In the demo this step is provisional: the detection model is still to be specified.*
  3. Draft
    A small AI drafts a strengths-based report from the evidence — nothing invented. Every draft is grounded on the child's own records and cites its sources. Missing data is shown, not filled in. Wording follows an approved, respectful terminology guide.
  4. Approve
    The teacher edits and approves every word before anything is shared. A rejected suggestion is discarded — nothing is saved until the teacher accepts. Accepted text carries the label 'AI-assisted · teacher approved'.
  5. Share
    With consent, the report reaches the family — clearly marked AI-assisted, human-approved. Nothing reaches a family until a parent gives consent. Under partial consent, video stays with the school. If consent is withdrawn, the family view empties at once, with a plain explanation.
  6. Oversee
    Programme, NGO and researcher see de-identified figures, with a full audit trail. Groups of fewer than k children are hidden and shown as 'cohort < k', in charts as well as tables. Every AI call, consent change and share is recorded — by student id, never by name.

What it does not do

These are design choices, not gaps. They are built into the platform — and the whole AI layer can be switched off, feature by feature, while the portfolio keeps working.

  • It suggests — it never decides. — The AI never diagnoses, labels or scores a child, never sets a support tier, never decides eligibility and never starts an intervention. Its wording stays at 'possible association' or 'example'.
  • Nothing it writes reaches a student or parent until a teacher approves it. — No report or note is ever sent automatically. Every AI output can be understood, corrected, rejected and traced.
  • No scores, grades or rankings — anywhere. — Progress is measured against the child's own past, never against other children. There is no field for a diagnosis, a label, an eligibility decision or a risk score.
  • It does not act on welfare concerns by itself. — If the platform brings up a concern about a child's wellbeing, it goes by a set route to the responsible person at the school — the AI never acts on it. In the demo a teacher looks at each 'needs a look' item and decides; the decision is recorded in the audit trail.
  • It gives a child no path to sensitive data. — The child view is read-only and strengths-only: no inputs, no AI, no numbers that read like grades, no labels, no alerts, no child photos in lists — and always with an adult.
  • It never lets one school see another. — Each school's data is kept separate, and access is closed by default and opened only by role. Attempts to cross schools or roles are blocked and logged. The school admin cannot open a portfolio; NGO and research views never show names or media.
  • Every AI action is logged. — Each output records the model, prompt and version used, and every AI call appears in the audit trail.

Where things stand

What you see here is a demonstration. It uses synthetic data only — no real children — and every figure on screen comes from sample data. It is not in use in any school, and it is not the programme's official website.

A few things are still being decided. Where they appear, the page says *to be confirmed*:

  • the parent and caregiver consent model — the demo shows the delivery team's design;
  • whether 'student' means neurodivergent pupils only or all pupils, and whether pupils get their own accounts;
  • the detection model behind clip signals — the demo shows provisional signals;
  • the offline capture path, which is not yet designed.

No dates, school counts or results appear here, because none is confirmed.