P

STUDENT

Blessing Masukwedza

SUPERVISOR

Mansha Nawaz

COURSE

BSc Information Technology, Teesside University

01

The problem and my research question

02

What Physiq is and its safety checks

03

Research and investigation

04

Method and evaluation

05

Ethics, legal issues and the project plan

10m+

People in the UK with arthritis

Many plans rely on exercises done alone at home.

2.4bn

People who may benefit from rehabilitation

A global need, far larger than clinics can supervise.

Short

Term gains in adherence

Reviews show apps help mainly in the short term. Evidence is limited.

Unsuitable advice

Language models can suggest exercises that are wrong for some conditions. Reviews report contraindicated recommendations.

Prompts are not enough

General chatbots show mixed accuracy on rehab questions. Even safe-looking answers can mislabel intensity.

Humans must stay involved

Professional guidance warns against over-reliance without clinician oversight.

SAFETY

Fewer contraindicated suggestions and clearer red-flag signposting.

ACCESSIBILITY

Usable for disabled adults and people with long-term conditions.

USEFULNESS

Helps people follow sessions and understand guidance.

BOUNDARY

Only programmes already set by a physiotherapist or doctor.

Cross-platform

Built to run on iOS, Android and the web.

Clinician-set programmes

Supports home exercises a physiotherapist or doctor already prescribed.

Simple, secure design

A small serverless proxy handles AI requests, so keys stay hidden. Health data stays on the device, with export and delete.

A clear boundary

Guidance and tracking only. It does not diagnose or prescribe.

01

Rule-based screening

Exercises are checked against each condition’s restrictions and avoid-lists. Unsafe ones are blocked or replaced.

02

Output validation

The AI must answer in a defined, structured format. Invalid or missing fields are rejected before they reach the user.

03

Red-flag escalation

Reported red-flag symptoms point to NHS 111 or a GP. Human support comes first.

Start from a programme

Pick a template for five rehab areas and edit it to match what your physio prescribed.

Track real sessions

Log what you completed and when. Record pain over time to spot patterns.

Own your data

Records stay on the phone, with export and delete options.

OPTIONAL EXTRAS

After the essentials

My scans

View the scans your physio gave you, with their notes. Physiq never reads or explains a scan.

Meals library

Simple recipes with steps, and optional tracking. You choose, Physiq does not pick a diet.

Real users first

Key journeys for disabled adults and people with long-term conditions. Vision, motor and fatigue needs all count.

Practical features

Screen-reader labels on buttons, inputs and messages. Large touch targets. Text that scales without breaking the layout.

A standard to meet

WCAG 2.2 AA guides decisions and acceptance checks. It is a core deliverable, not a bonus.

Clinical and user needs

Why people struggle to keep up home exercise. Safety limits, contraindications and red flags per condition.

Guardrails for AI

Rule checks on outputs, structured-output validation, and resilience to prompt injection.

Accessibility and rivals

WCAG 2.2 AA and React Native patterns. Competitor apps such as Accessercise, Budy and Kaia, to find gaps in safety and inclusion.

Mixed methods

A short questionnaire on home-exercise needs, then a few interviews to explain motivations and barriers.

Iterative build

Two-week solo Scrum sprints with a prioritised backlog. MoSCoW keeps the essentials first.

Tested throughout

The safety layer is the most test-critical part. AI output is checked against its required format continuously.

Fits the deadlines

Handles changing requirements while meeting fixed academic dates.

Safety benchmark

Run the same scenarios with the rules on and off. Compare how often each version gives unsafe guidance.

Synthetic test set

160 fictional personas and 494 scenarios across the rehab areas. The test kit is built and is still to be run and reviewed.

Usability testing

Think-aloud sessions to find real interaction problems, and the SUS questionnaire, reported as a reference score.

Overall judgement

Safety results and usability findings are combined to judge usefulness. Limits are reported honestly.

Approval first

No participant involvement until full School Ethics approval.

Low-risk participation

People only evaluate the app. They are not asked to exercise or change any treatment.

Synthetic data by default

Only fictional personas and scenarios go to the AI. Demos use made-up data, and scan pictures never leave the device.

Privacy and withdrawal

Accessible consent materials, withdrawal at any time, and the option to decline AI features.

Privacy

Physiq handles special category health data, so privacy is built in. Data stays on the device, with export and delete.

Medical device boundary

It supports clinician-set programmes and does not diagnose or prescribe. Wording and features keep it outside diagnostic claims.

Professional duty

Follows CSP guidance on consent, transparency and human escalation, and the BCS duty to protect wellbeing and reduce harm.

SEMESTER 1

Foundations

Research, requirements and early design. Ethics approval before any questionnaire, interview or testing.

SEMESTER 2

Build and prove

Build the app and backend, test safety continuously, run the benchmark and usability study, then write up.

DELIVERABLES

What I hand in

A cross-platform prototype with its serverless proxy, a safety and usability evaluation, and a weekly public project blog.

Measure it in a controlled before-and-after benchmark.

Deliver core journeys disabled adults can use.

Keep health data on the device and AI behind a secure proxy.

Stay support software, never a diagnosis or a clinician replacement.