This page shows how I designed Physiq: my process, the decisions I made and why, the accessibility choices, and how the screens changed when I moved from fitness to physiotherapy.
01
Discover
Literature review and app review. Survey and interviews only after ethics approval.
02
Define
Personas, problem statements and a prioritised feature list.
03
Ideate
Sketches and user flows for the programme, the pain check and the coach.
04
Prototype
I drew wireframes first, then built it in Expo / React Native. This page also has an early browser demo.
05
Test
Planned: safety checks on synthetic test cases, a usability test and a review by physiotherapists.
Design decisions and the evidence behind them
Each safety choice below links to something I read. These are design reasons, not proof that the design works. Testing will show that.
The programme comes from a clinician
Physiq does not choose exercises or change a programme. It supports the one your physiotherapist or doctor gave you.
Why: Authors of the LLM studies say these tools should stay supplementary and supervised (Safran and Yildirim, 2025; Cali et al., 2026).
Rules check the AI
A rule-based layer filters the coach’s replies against restrictions and an avoid list before you see them.
Why: Models can score well on safety and still get intensity wrong, and the authors call for rule-based checks (de Mendonça, Sousa and Coelho, 2026).
Red flags stop the conversation
If you mention a possible red-flag symptom the coach stops and points you to NHS 111, your GP, your physiotherapist or 999.
Why: Accuracy varies by topic and by question (Scaff et al., 2025), so serious symptoms should never depend on a model getting it right.
Extra care after surgery
Post-operative cases get cautious wording and a clear pointer back to the surgical team or physiotherapist.
Why: Expert raters found the weakest AI programmes in timing-sensitive cases such as ACL reconstruction (Cali et al., 2026).
Your data stays on your device
Pain and exercise logs are stored on the phone and not sent to a server for storage.
Why: Health data is special category data under UK GDPR (ICO, no date).
Support, not a medical device
The wording says Physiq helps you follow your programme. It does not diagnose or treat.
Why: MHRA guidance ties the status of software to its intended purpose (MHRA, 2023). I am not claiming any regulatory approval.
Accessibility
Accessibility is a design principle, not an extra. Some of this still needs testing.
01
Your area comes first
The area your programme is for is part of the setup, not hidden in settings.
02
Bad days are allowed
Flare-up check-in pauses sessions and points to your physiotherapist’s advice. A bad day is not counted as a failure.
03
Big targets, calm visuals
Tap targets aim for at least 44px, with a quiet colour palette. Contrast and screen-reader checks against WCAG 2.2 AA are still to be done.
04
Less typing
Quick actions and one-tap choices reduce typing. Testing with assistive technology is planned.
05
Plain-language coaching
No jargon and no shaming. The coach writes the way a supportive friend would.
Wireframes: how the screens changed
The app changed direction part way through, so I am showing the wireframes in four stages. I redrew each one in low fidelity from the screens as they were at that point. Real screenshots of the old and new versions are further down this page.
Stage 1 · Where it started
Fitness first
I began with an adaptive fitness and nutrition app. Set-up asked for 14 things, Today was a calorie ring, and Train made a weekly plan and moved people through rehab phases. I redrew these four screens from the version I tagged fitness-v1.
What I found wrong: it made plans and progressed rehab phases, which is a clinician's job.
Set-up
14 steps, step 1
Today
Calories, macros, steps
Train
Generated plan and rehab phases
Coach
Fitness chat
Stage 2 · Early October
Safe and accessible fitness
Before changing direction I made the existing app safer. Messages are checked for emergency wording before the AI sees them, and replies are checked before I show them. I also added accessibility settings, read-aloud for workouts and sign-in.
What stayed: the safety checks, the accessibility settings and read-aloud carried into the physio version.
Coach safety check
Red flag stops the AI
Accessibility
Text size, contrast, motion
Hands-free workout
Read aloud, set by set
Sign in
Email and password
Stage 3 · 15 to 17 October
The physio direction
I decided to focus on people following a home exercise programme from a physiotherapist or doctor. The first physio wireframes ask only for a name and what the programme is for. The person types in their own exercises from the sheet. Physiq does not choose them.
What changed: no plans from the app, no rehab phases, no streaks. Pain is something I record and share with the physio.
Set-up
Two questions
Programme
From the physio sheet
Today
Programme, pain, flare-up
Progress
Pain chart, no streaks
Stage 4 · Where it is now
The Expo app rebuilt
I rebuilt the real app around the same flow. A session has a stop button for worsening pain. The Helper answers general questions only, and emergency or crisis wording gets a fixed safe reply with no AI call. Meals, steps and weight are still there as an optional extra, switched off by default.
Still to do: test it on a phone with a screen reader and with real people once ethics is approved.
Set-up
Name and area
Today
Programme and pain check
Session
Stop button
Helper
Safe replies
Settings
Fitness extra off
The app, before and after
These are real screenshots from the web build of the Expo app, taken on 17 October 2026 with made-up test data for a person called Sam. I have not captured them on a phone yet. The old screenshots needed a one line fix, because the old app crashed on web with an infinite re-render error. I kept the old version unchanged apart from that fix.
Before
Fitness first (fitness-v1)
Set-up, 14 stepsToday with calories and stepsTrain, generated plan and rehab phasesFitness coach
Now
Physio first
Set-up, two questionsToday with programme and pain checkSession with a stop buttonProgress, no streaksHelper safe reply, no AI callSettings and fitness extra
