Inspiration
Families caring for an aging parent end up with medical papers scattered across clinics, phones and WhatsApp. At every appointment the caregiver is asked "what is she taking now?" and answers from memory. In an emergency, nobody else in the family can find anything. I wanted a private, organized record that stays on the family's own NAS.
What it does
Elder Care Record & Visit Brief has three modes:
- Intake: lists the documents in a folder, classifies them (prescriptions, labs, discharge summaries, appointment letters) and after confirmation, copies them into an organized, tagged care-record folder. Originals are never moved or deleted.
- Ask: answers questions about the record, such as "what dose was she prescribed?", with the source file and page. If the record doesn't contain the answer, it says so.
- Brief: builds a doctor-visit brief and an emergency sheet with current medicines, places where documents differ from each other, follow-ups with no appointment, changes since the last brief, questions for the doctor, estimated refill dates and the caregiver's voice-note instructions. If the parent has agreed, it also compares home camera events against a baseline period (counts and times only).
It never gives medical advice. Differences are flagged for the doctor or pharmacist.
How I built it
The Skill is an orchestration workflow built with the UGOS Skill Creation Tool. It delegates in order to the NAS's own experts for files, documents, knowledge and monitoring. Each step uses only the real result of the previous one and every change needs user confirmation. It uses no external services or models.
Challenges
- Keeping it safe: no advice, no interpretation, no guessing from unreadable files.
- Making the camera step consent-gated and limited to counts and times, with no identification of people.
- Staying inside the supported NAS capabilities, which ruled out ideas such as reading medicine strips from photos.
What I learned
Trust is the feature in a health-adjacent tool. Source links, honest "not found" answers and confirmation before every change matter as much as the extraction itself.
What's next
Round 1 is an early build, not yet run on a NAS. Next I would test it with synthetic documents: extraction accuracy, scanned files, the copy-and-tag flow, source-linked answers, camera filters and the previous-brief comparison.
Built With
- ai
- creation
- markdown
- nas
- skill
- skill.md
- tool
- ugos
- ugreen
