Ubat & Rekod

• 

  ugos

Inspiration

What it does

How we built it

Challenges we ran into

Accomplishments that we're proud of

What we learned

What's next for Ubat & Rekod

Inspiration

In many Malaysian families, adult children look after parents who see two or three clinics. The paperwork ends up as phone photos spread across siblings, and those siblings do not all read the same language. Answering "what is Ma taking right now?" or "is she getting the same medicine from two places?" means digging through chats. Health records are also among the most private things a family owns, so they should stay at home.

What it does

Ubat & Rekod ("medicine and records") turns a UGREEN AI NAS into a private family health binder.

• Anyone in the family drops photos or PDFs of prescriptions, pharmacy labels, appointment cards and lab results into one Inbox folder.
• The Skill reads them with on-NAS OCR, vision and document experts. You confirm anything uncertain, and it files and tags each record by person, type and month.
• It keeps a dated medication list, timeline and alerts for each person. Alerts cover possible duplicate medicines across clinics or brand names, strength changes, refills running out, missed or upcoming appointments, and lab values the lab itself flagged.
• Ask in chat: "When is Ma's next appointment?" You get the date, the preparation needed (for example fasting) and the source file.
• It translates summaries for each sibling (Bahasa Melayu / 中文 / English) and drafts a one-page brief for the next doctor visit.
• It can also sweep the photo library for old clinic slips and copy them in.

Every line cites its source document. Every alert is phrased as a question for the doctor or pharmacist. It is not medical advice.

How we built it

A pure SKILL.md orchestration that follows the Uliya Skill Development Standard. There are no scripts. The parent Skill delegates single-purpose steps to four registered UGOS experts:

• file-agent: get-user-dir, find, sha256sum, scan-dedupe, mkdir, cpmv, rename, file tags
• photo-agent: vision-ocr, vision-vlm, search-screen, search-picture
• document-agent: extract with the medical-report schema, reconcile, qa, translate, writing
• knowledge-agent: save-text, collect-evidence

Read steps run freely. Every move, rename, tag, save or translation is shown as a plan first and goes through gateway or workflow approval. Nothing is ever deleted. Experts are asked to use the NAS local model, and if only a cloud model is available the Skill stops and asks. The package has seven reference files: the extraction prompt and schema, the binder layout and naming rules, a trilingual glossary, the visit brief template, safety rules, a fictional test plan and the disclosures.

Challenges we ran into

• Keeping a health tool useful without crossing into medical advice. We solved this with question-style alerts, never inventing unreadable values, and "not recorded" instead of blanks.
• Many pharmacy labels and appointment cards carry no patient name. We added person matching with a confirmation step.
• Malaysian DD/MM dates, mixed-language slips, and brand names versus active ingredients.

Accomplishments that we're proud of

• An end-to-end flow from photos to filed records, a medication list, alerts, a translation and a visit brief, all on the NAS.
• Explicit handling of every delegation state (completed, failed, cancelled, waiting_user, waiting_approval) plus empty and partial results.
• The package passes the Uliya Skill Kit static checker with zero errors and zero warnings.

What we learned

The NAS is the right home for this kind of data. Local storage, local AI and the family's own folder permissions solve the privacy problem that stops families from using cloud chatbots for medical paperwork.

What's next

• Validate the fourteen fictional test cases on the assigned NAS, if selected as a finalist.
• Tune the extraction prompt for handwriting and for faded thermal labels.
• Add reminders once UGOS offers a scheduling capability.

All sample people, clinics and brands in the package are fictional. No real health records were used.

Built With

• ugreen-ai-nas