Family Medicine Cabinet

Inspiration

I worked for about 20 years as a pharmacy assistant in community pharmacies in Bucharest. Again and again, people brought bags of old medicines to the counter, or asked me which boxes at home were still in date. Checking a medicine cabinet box by box is a chore everyone puts off, and photos of your medicines are private health information that most families would rather not send to a cloud service.

A home NAS is the natural place for this: the photos stay with the family, and the AI comes to the data instead of the other way round.

What it does

Family Medicine Cabinet is an AI Skill for UGREEN AI NAS. The family photographs their medicine boxes into one NAS folder. The Skill:

• lists the photos in that folder;
• reads the printed text on each box: the product name and the expiry date exactly as printed;
• sorts every item into Expired, Check soon (within 30 days), 31 to 90 days, Later, or Date unclear;
• makes a pharmacy take-back list of the expired items;
• shows a preview and, after the user agrees, saves a tidy note on the NAS.

It never gives medical advice, doses or interactions, never says what a medicine is for, and says plainly that it can misread labels. Unclear dates are listed apart, never guessed.

How we built it

I built it as a solo entrant with Claude Code, using the official uliya-skill-kit (0.7.3) provided by UGREEN. The Skill is a Markdown orchestration Skill that delegates work to three NAS experts in order:

1. ug-filesystem (file-agent) lists the photos;
2. ug-photo (photo-agent) reads the printed text;
3. ug-knowledge (knowledge-agent) saves the note, with platform approval and never overwriting.

I wrote and approved every medicine-related sentence myself, using my pharmacy experience: for example, month-only dates count until the last day of that month, and a separate line reminds people that eye drops, syrups and creams must be thrown away a set time after opening.

Challenges we ran into

• Keeping the AI honest: the Skill must show only what is printed and never "fill in" a smudged name or a two-digit year.
• Privacy: I removed prescriptions and receipts from the scope, so the Skill reads only medicine box photos.
• I don't own a NAS yet, so everything so far is checked statically, not on real hardware.

Accomplishments that we're proud of

• A Skill that passes the kit's static check with no errors or warnings.
• A 26-case synthetic test sheet covering date boundaries, unclear dates, a label with a patient name, a photo that tries to change the rules, empty and missing folders, and medical questions the Skill must decline.
• Wording that a pharmacist would sign off on: careful, clear and never alarming.

What we learned

The hardest part of an AI health tool is deciding what it must never say. A narrow, honest Skill is more useful than a clever one that guesses.

What's next for Family Medicine Cabinet

If I become a Finalist, I will test text recognition on real packaging (glossy, curved and embossed boxes), the approval prompt and the saved note on a real UGREEN AI NAS, then record the demo.

Built With

• claude-code
• markdown
• ocr
• ugos
• ugreen-ai-nas
• uliya-skill-kit