One Page for the Doctor

Inspiration

In Filipino families, one person usually becomes the caregiver who takes Lola to her checkups. Her records pile up over the years: lab PDFs from three hospitals, phone photos of prescriptions, a discharge summary somewhere in a folder. In the clinic, nobody can remember her last creatinine or which Losartan dose is current, and siblings abroad keep messaging to ask what the doctor said.

What it does

One Page for the Doctor is a Uliya Skill for UGREEN AI NAS. One request, like "Prepare Lola Nena's one-pager for her cardiology checkup on Friday", produces a printable visit brief with:

• current medicines, and any dose that changed between prescriptions
• lab results as trends across visits, with the report's own high and low flags
• allergies and recent hospital stays
• factual points to raise with the doctor (a changed dose, a paper that could not be read)
• a source for every value: file name and page

The caregiver can then ask follow-up questions, like "What was her potassium in March?", and get the exact value with its source. The Skill only gathers and cites. It never interprets results or suggests a medicine or dose.

How we built it

The Skill is a single SKILL.md with two linked references (record rules and brief format), built with the official Uliya Skill Kit. It delegates to four NAS experts:

• file-agent: finds documents, creates the visit folder, copies the sources (never moves or deletes originals)
• photo-agent: finds photos of paper records by the text in them, then reads them with OCR and the vision model
• document-agent: extracts values with the medical-report schema, reconciles prescriptions, answers follow-up questions with page evidence
• knowledge-agent: transcribes an optional recording of the doctor's instructions and saves the brief

Challenges we ran into

Health data leaves no room for guessing. The Skill shows conflicting values side by side instead of choosing one, stops to ask when two patients have similar names, and writes "could not read" rather than guessing at handwriting.

Accomplishments that we're proud of

The package passes the official check-skill.mjs with 0 errors and 0 warnings, and handles all five delegation states.

What's next

Testing on a real UGREEN AI NAS with synthetic lab reports, prescriptions and a recorded doctor's instruction, then tuning extraction for common Philippine lab report layouts.

AI disclosure

Idea and direction are mine. I used AI tools (Claude, and another AI chat assistant) to strengthen the idea and to help draft and check the Skill files and proposal.

Built With

• markdown
• ugos
• ugreen-ai-nas
• uliya