Inspiration
Clinic quality coordinators and research administrators often inherit scattered controlled documents. Identical document codes can carry conflicting versions or effective dates, making manual pre-circulation checks slow and error-prone.
What it does
AI Clinical Document Guardian is a privacy-first, read-mostly Uliya Skill for UGREEN AI NAS. It reviews up to 12 explicitly authorized non-patient SOPs, quality policies, and research templates; flags conflicting revision metadata and duplicate candidates; and produces a source-linked Clinical Document Integrity Ledger for human review. It never determines clinical correctness or edits original documents.
How we built it
The Markdown Skill delegates discovery and duplicate scanning to file-agent, metadata extraction and reconciliation to document-agent, and optional approved ledger archival to knowledge-agent. It uses returned source identifiers, explicit evidence coverage, and CONSISTENT, CONFLICT, NOT_VERIFIED, DUPLICATE_CANDIDATE, and INACCESSIBLE verdicts. Local NAS processing is required; no external runtime APIs, libraries, datasets, or medical records are bundled.
Challenges and lessons
Missing provenance must remain NOT_VERIFIED. A metadata mismatch is not proof that either document is wrong. The workflow handles completed, failed, cancelled, waiting_user, and waiting_approval states and requires approval before optional saving.
Current status
This Round 1 entry contains a drafted Markdown Skill, supporting contracts, disclosures, a prospective test plan, and a one-page proposal. Independent local structural checks were completed. Official creation-tool validation, supported expert-action compatibility, local-model behavior, and NAS execution remain unverified. No NAS demo or performance results are claimed.
What is next
Validate against the official creation tool and assigned NAS using fully synthetic conflicting and consistent SOP pairs, a duplicate file, and empty/unreadable/privacy edge cases. Verify provenance, approval-before-save, and unchanged originals.
Third-party disclosures
UGREEN UGOS/Uliya is the required proprietary target runtime under UGREEN terms: https://ai.ugreen.com/pages/hackathon-starter-kit . OpenAI ChatGPT was used for design, research, drafting, and local structural checks under OpenAI terms: https://chatgpt.com/ . It is not a runtime dependency. No third-party code or datasets are bundled.
Document-control review only; no medical, legal, or regulatory validation.
