Before
Documents, certificates and non-conformities were distributed across shared folders and part of the review meant checking files one by one. Expired information or a hard-to-find pending item made audits and follow-up harder.
QUALITY · DOCUMENT AUTOMATION
TECHNICAL PILOTCentral ISO
Traceability, deterministic rules and evidence
The project started from conversations with Quality stakeholders to test how much of a shared-folder, manual-checking process could become traceable verification without changing official documents.
Validation32 passing tests
ControlSHA-256 · change-based reprocessing
StateFunctional PoC · not certified
The pilot
The scanner reads the source, hashes what it finds, reprocesses only changed files, extracts content locally and sends structured data to a Rules Engine. The interface surfaces pending items and evidence, while the final conclusion remains with the Quality owner.
Explainable rules
Critical rules remain deterministic so each alert can be explained. AI may support retrieval or summarization in future iterations, but it is not used as a black box for compliance decisions.
Security by design
Read-only document source, synthetic data in the public edition, no real credentials in Git, local processing by default and human review for sensitive decisions.
Architecture
Document repository → Python scanner → local extraction → SHA-256/classification → SQLite/PostgreSQL → Rules Engine → FastAPI → dashboard/Tauri. n8n schedules queries and alerts through the API.
Real scope
I do not claim ISO certification, automatic compliance, corporate SSO, running RAG or official production deployment. This is a functional technical pilot.
The value is not simply “reading PDFs”; it is being able to show which rule created a pending item, which evidence supported the alert and keeping the official document source outside the pilot's write path.
Next projectCompass UOL Program
Contact