How it is going

Adding scans and a meals library

My supervisor showed me what master's students are doing and had an idea: people could see their own physio scans in the app and understand what is going on. I like the idea, but I have to be careful. An app that tells someone what a scan shows is diagnosing, and Physiq must not do that. It would also need machine learning, which I am not supposed to be doing.

So I built the safe version in the web demo:

• Scans tab. You add a picture of your own scan, zoom in, change the contrast, and write down what your clinician told you and your questions for the next appointment. The app does not read the scan. The demo uses a clearly labelled illustration, not a real scan, and pictures stay in the browser.
• Meals tab (optional extra). A recipe library with ingredients, cooking steps and rough calorie examples. Filters like vegan, gluten free, higher protein or low effort let you match the diet your physiotherapist, doctor or dietitian gave you. Physiq does not choose the diet. Food, weight and activity tracking sit in the same tab.

The calorie figures are my own rough examples and are not checked against a food database yet, and I say so in the app. Next I need to source proper values, check what ethics and data protection need for health images, and decide how to bring both features into the Expo app. A later idea is letting a physiotherapist suggest recipes, but I have not built that.

My supervisor is happy with the site and the app

My supervisor said it is fine to have the app on the site, which I was not sure about, so that is a relief. Her advice now is to keep making it better and to build up the blog.

So the plan is to add the rest of her list (tracker below), rewrite the research, and write the blog as I go instead of at the end. The blog will also help me when I write the report, because I can see what I decided and why.

My project proposal

This is the written proposal for my final year project, due on Friday 30 October. It covers the research question, the objectives, the background research, ethics, the method and the plan. It is a working copy, so I will replace it with the final version after I hand it in.

My proposal slides

These are the slides for my project proposal. The main aim is to test whether rule-based safety checks reduce unsafe AI exercise guidance. Scans and meals are optional extras that only get built after the must-haves.

Supervisor feedback tracker

Every point from my supervisor, and where it is now.

From fitness app to physio tool

The project went through three stages, and the Design page shows the wireframes for each one.

• Fitness first. An adaptive fitness and nutrition app with a calorie ring, a generated weekly plan and a fitness coach.
• Safe and accessible fitness. Before changing direction I added emergency wording checks, accessibility settings and a hands-free workout.
• Physio first. The programme comes from the physiotherapist, and Physiq helps the person follow it. The safety checks and accessibility work carried over.

Looking at the three side by side, the big lesson was that my first version made plans and moved people through rehab phases. That is a clinician's job, not an app's.

Wireframes: how the screens changed

I drew low-fidelity wireframes in four stages, from the fitness-first start to the physio app I have now. Putting them side by side showed me what I got wrong at the start: the first version made plans and moved people through rehab phases, which is a clinician's job.

Rebuilding the app around physiotherapy

I rebuilt the Expo app so the programme comes first. Set-up asks for a name and what the programme is for. Sessions have a stop button, and the Helper gives a fixed safe reply if someone mentions an emergency or a crisis. I kept the old version in an archive folder so I can compare them. I have only tested it on the web so far, not on a phone.

What I have written so far

Alongside the app I have been writing and checking the research. The site now has a sourced problem section and a summary of the literature, and it says what each source can and cannot show. Survey and interview work is planned, not done, and it waits for ethics approval.

I also worked with a database pack from my lecturer and added my own made-up test data, 160 personas and 494 test scenarios. It has not yet been run in SQL Server, and a physiotherapist has not reviewed it. I am being careful not to claim more than I have actually built, tested or measured.

Why I changed direction

I changed Physiq from a fitness app to a tool for people following a home exercise programme from a physiotherapist or doctor. It does not diagnose, prescribe or replace clinicians. Accessibility is a design principle.

Week one: the first draft of my project proposal

Week one of semester one was the week of 28 September 2026. The job was to turn my Physiq prototype into a project I could propose. This post is what I did and what I got wrong. I wrote the first draft for a fitness app, and I have kept it as it was so I can show how the project changed.

Where I started

Physiq was already a working prototype in Expo and React Native. It was a fitness and nutrition app with an AI coach, for adults with disabilities and long-term health conditions. My first working title was Designing and Evaluating a Safe, Accessible, AI-Assisted Fitness and Nutrition App for Adults with Disabilities and Long-Term Health Conditions.

Why I thought it was needed

Activity Alliance (2026) report that 39.3% of disabled adults in England were inactive in 2024 to 25, compared with 19.7% of non-disabled adults. I read that as a sign that mainstream fitness apps are not built for many disabled people. I want to check the figures again against the source before the final proposal.

My first research question

To what extent can a condition-aware mobile app that combines rule-based safety constraints with a large language model give exercise and nutrition guidance that is safe, accessible and useful for adults with disabilities or long-term health conditions?

I split it into three smaller questions: what users need, whether adding safety rules to an AI actually reduces unsafe advice compared with prompting alone, and how usable and accessible people and an expert find the result.

What I found when I audited my own code

Before writing anything I read through my own prototype, about 6,400 lines of TypeScript. The good parts were a broad model of conditions and a complete user journey. The problems were more important:

• The AI and API keys were in the app bundle, so anyone could extract them.
• Safety rules only existed as wording in the prompt, with nothing checking what the AI replied.
• The calorie adjustments per condition were not traced to any evidence.
• Health and pain data was stored on the device with no consent, export or delete controls.
• There were no automated tests, and the app used imperial units for a UK audience.

I have already fixed some of these since, such as moving the AI call to a server proxy and adding automated tests. The rest are still on my list.

How I planned to work

• Scrum adapted for one person, with two-week sprints and a prioritised backlog.
• A literature review and an app review first, then a questionnaire and interviews only after ethics approval.
• Evaluation with a safety benchmark on made-up personas, a review by a physiotherapist, and usability testing with the System Usability Scale. I will report the results rather than promise a score.

Dates I am working to

• Final proposal and ethics form: Friday 30 October 2026, 16:00
• Progress poster: 7 December 2026
• Artefact and walkthrough video: 18 March 2027
• Final report: 26 April 2027

What changed afterwards

On 15 October I changed the focus from fitness to physiotherapy. Physiq now supports people following a home exercise programme from a physiotherapist or doctor. This draft is for the old version, so the research question and objectives are being rewritten. The later posts explain why.

Reference: Activity Alliance (2026) Our response to Sport England's Active Lives 2024-25 report. Available at: https://www.activityalliance.org.uk/news/10563-our-response-to-sport-englands-active-lives-202425-report (Accessed: 30 September 2026).