Inspiration
Medical records from different hospitals often end up in one disorganized folder. Patients and caregivers need to find information and detect inconsistencies without uploading private documents to a cloud service.
What it does
Custodia organizes medical PDFs into a tagged, cross-referenced archive and creates a plain-text digest where every finding cites its source document and page. It checks discharge medications against prescriptions, identifies date inconsistencies, and distinguishes compared agreement from unreadable or unprocessed documents. Duplicates are grouped while every original file is preserved.
How I built it
I created a Uliya skill using the official UGOS Skill Creation Tool. The workflow delegates only to registered file, document, and knowledge experts. Tag changes, file moves, and digest saving follow approval states. Extraction and reconciliation use bounded processing and preserve actual returned coverage. Everything runs locally on the NAS.
Challenges I ran into
The main challenge was preventing a polished digest from implying evidence or coverage it did not have. Custodia now requires source-and-page citations, distinguishes unreadable from unprocessed files, and requires save-verification fields before claiming a digest exists.
Accomplishments
The complete skill package and four reference guides pass the official static checker with zero errors and zero warnings. The ZIP was independently extracted and checked again, and every response in the proposal PDF was verified against the rendered file.
What's next
Runtime NAS validation is still pending. As a Finalist, I will test medication reconciliation, date inconsistencies, unreadable image-only PDFs, approval states, and partial coverage on the assigned device.
Built With
- markdown
- ugos
- uliya
