PROJECT OVERVIEW
iNurra — a patient-first home-care tool
iNurra helps people understand and follow complex care instructions after they leave a hospital or clinic. It uses AI-assisted document and image analysis to turn the paperwork someone is already holding into a plain-language visual care plan.
It is a support tool, not a medical device. It does not diagnose. Everything it reads from a photo has to be reviewed and confirmed by the person before it becomes part of their plan, and anything urgent or unclear goes to a qualified healthcare professional.
Problem
Someone leaves the hospital holding a folder. Inside are several medications on different schedules, brand names sitting next to generic names, a follow-up appointment, and a line telling them to check their blood pressure every day.
They may also need to watch a wound, read documents written for clinicians rather than for them, and work out on their own when something is worth a phone call.
When those instructions are hard to follow, doses get missed and appointments slip. Some people end up back in hospital. Hospitals carry real cost and operational pressure from readmissions that could have been avoided. The goal was to turn that lose-lose into something better for patients, caregivers and care teams.
My Role
I have designed and built iNurra on my own since January 2026, and I am still working on it. I am not a clinician, so I framed the problem around what a person is physically holding when they get home rather than around clinical practice.
- Framed the problem from what patients are physically holding when they get home, rather than from clinical workflow.
- Interviewed patients and family caregivers, and used what they said to decide what the product would and would not do.
- Designed the flows, the information architecture and the screens, working inside an existing design system rather than building a visual identity.
- Wrote the plain-language content, including how the product explains a medication and how it phrases a safety prompt.
- Set the safety boundaries: no diagnosis, no silent writes, and a mandatory human review before anything reaches a care plan.
- Built it and took it through App Store submission, and AI-paired throughout with Claude Code.
Target Users
The product is built for people managing care at home, and for the people helping them. That includes older adults with complex medication routines, and people facing a language or health-literacy barrier.
The patient who just got home
Tired, often sore, sometimes reading through the wrong glasses. They need one question answered before anything else: what do I take, and when.
The family member helping
Usually an adult child, often managing care remotely. They want to know what needs attention without reading every page of a discharge summary.
The care team
Clinicians who would benefit from earlier signal when something at home is going wrong. The provider-facing side is a concept and is not built.
User Research
I spoke with patients about managing medication and home-care instructions. I interviewed seven patients and four family caregivers who had recently managed a discharge plan or a long-term condition. I asked each of them to show me how they were keeping track, and to read a piece of their own paperwork aloud.
I know I take the white one in the morning. I couldn't tell you what it's for.
There's a note on the fridge, three alarms on my phone, and the bag from the pharmacy. None of them quite agree.
I take a photo of her foot every night. I still can't tell you if it's better than last week.
Key Insights
Three findings changed the product more than anything else.
Confusion, not refusal
Nobody I spoke to had decided to skip a dose. They could not always tell two white pills apart, or turn a written instruction into a time of day. The job was to translate, not to nag.
One routine, spread across four places
People were using paper notes, phone alarms, a calendar and the medicine bottles at the same time. Nothing reconciled them, so the routine broke wherever those four disagreed.
Caregivers want the summary, not the document
Family members were not short of information. They were short of a way to see what needed attention without reading a clinical document end to end.
Design Decisions
Three decisions did most of the work, and each one carried a cost.
Start from what the person already has
01
Asking someone to type in every medication, appointment and task is the fastest way to lose them on day one. So the product starts with a photo of the label, the discharge sheet, or the monitor. The cost is that the app now has to be right about a curled, badly-lit picture, which is a much harder problem than a form.
Nothing is saved until the person confirms it
02
The AI reads the photo and proposes. The person reviews it item by item and corrects anything wrong before a single thing is written to their plan. It costs the effortless demo and adds a step to every upload. It is also the only reason I would let this near a medication schedule at all.
Say what it means, not what it says
03
A reading of 148 means nothing on its own, and neither does a brand name. So the product explains what a medication is for and what range a number sits in. It never offers a diagnosis, and it names the point at which a person should stop reading and call someone.
Design Solutions
Each part of the product answers something a patient or a caregiver actually described.
Visual care calendar
01
Medications, appointments and daily care tasks land on one timeline. Scanning a discharge document surfaces possible appointments, schedules and tasks, and the person confirms them before anything is added. This is the answer to a routine spread across a fridge note, three alarms and a calendar.
Medication scanner
02
Photograph a bottle or an instruction sheet. The product identifies the likely details, explains them in simpler language, and proposes a schedule from the dosage frequency. The person reviews and corrects before saving. Moving one dose can shift the rest while preserving the required interval between them.
Document summaries
03
Upload a discharge summary or a care instruction and get a plain-language summary: the medications, the appointments, the tasks, and questions worth asking a provider. The original document stays available, because a summary should never be the only copy.
Body and skin tracker
04
Photograph a wound, a foot or a rash over time and see the images side by side. It answers the caregiver who takes a photo every night and still cannot tell whether it is improving. It does not diagnose. It helps someone document change and decide when to make a call.
Health measurement tracking
05
Record readings from home devices such as a blood-pressure monitor. The product organizes them over time so a pattern is visible, and so the history can be shared with a provider rather than remembered.
Safety prompts
06
When something in a photo, a document or a reading looks like it may need attention, the product recommends contacting a healthcare professional. It states plainly that a prompt is not a diagnosis. With the person's permission, a high-risk prompt could also reach a connected caregiver or care team.
Ask Nurra
07
A text and voice assistant that answers from the person's own uploaded documents, medications, appointments and plan. When is my next appointment. Which medication is the evening one. What did my discharge notes say about the wound. It names the document behind its answer, and sends anything urgent to a professional or emergency service.
Language and accessibility
08
Plain language, readable type, clear icons, voice input and a simple visual hierarchy throughout. Language support matters most for families where the paperwork is in one language and the patient reads another. Multi-language support is designed but not built.
The marked step is the one I would not automate. Everything before it is a suggestion.
Tradeoffs
Most of these are unresolved on purpose. In a health context the safe answer is usually the slower one.
AI speed against AI error
01
Reading a photo removes a great deal of manual work, and it can also be wrong. That is why review is mandatory rather than optional, and why the person sees each extracted item on its own card rather than as a finished plan.
A simple calendar against a complete instruction
02
A visual timeline makes a complex routine legible. It can also drop the parts that matter most: the interval between doses, whether something is taken with food, a warning attached to a medication. Simplifying the view must not simplify the instruction.
Photo tracking against privacy
03
Wound photos and medical documents are among the most sensitive things a person owns. The product needs explicit consent, secure storage, controlled sharing and a real way to delete everything. Anything less makes the feature not worth having.
Early warning against alert fatigue
04
Prompting someone to call a professional can help them act sooner. Too many prompts create fear, and then they get ignored. The product has to prioritize what is genuinely worth a call and say clearly what to do next.
Helpfulness against clinical authority
05
A missed dose cannot always be safely moved to a later time. Without verified clinical rules the product records the miss and gives general next-step guidance, and it does not reschedule on its own. Being less helpful is the correct behaviour here.
Impact
Observed or supported
iNurra was designed, built and taken through App Store submission by one person. The safety model holds in the shipped product: no model output reaches a care plan without a person confirming it first.
In early concept testing, people responded well to seeing medications, appointments and home-care tasks together on one timeline rather than in separate lists.
Expected
The product aims to help people understand their care plan and feel more confident managing it at home, and to help caregivers and care teams notice a problem earlier.
Reduced thirty-day readmission is a goal of this product, not a result it has shown. No clinical outcome has been measured, and no such claim should be made until real evidence supports it.
What I would track next
- Medication-plan comprehension
- Scheduled-dose completion
- Follow-up appointment attendance
- Completion of home-care tasks
- How often people correct an AI suggestion
- Caregiver engagement
- Patient confidence
- Support escalation rates
- Thirty-day readmission rates