Inspiration
Modern professionals and SMBs generate an unprecedented volume of sensitive financial records, bilateral contracts, and regulatory filings. While cloud-based LLM architectures offer document intelligence, they demand an unacceptable trade-off: streaming confidential corporate data to third-party endpoints. With the launch of UGREEN AI NAS and the UGOS ecosystem, we envisioned an uncompromising alternative—an autonomous, local-first intelligence skill that provides deterministic document auditing and semantic retrieval entirely on-premise, operating strictly air-gapped without leaking a single byte.
What It Does
Doc-Privacy-Auditor transforms UGREEN AI NAS into a proactive, autonomous compliance desk:
- Autonomous Filesystem Ingest: Listens continuously to designated storage pools via native Linux filesystem hooks (
inotify), acquiring atomic read locks upon file arrival. - On-Device Parsing & OCR: Extracts layout-aware textual layers and tabular datasets on the local SoC/NPU without cloud dependencies.
- Deterministic Validation Gate: Applies mathematical and schema invariant checks (e.g., verifying arithmetic identities where $\text{Subtotal} + \text{Tax} \equiv \text{Total}$, checking mandatory metadata, and detecting anomaly thresholds).
- Automated Routing & Quarantine: Validated files are tagged and moved to structured directory trees (
/audited/YYYY/MM), while anomalous or corrupted documents are isolated in/quarantinealongside structured machine-readable error logs. - Local Knowledge Indexing: Builds privacy-preserving local semantic embeddings for instant search across authorized internal network shares.
How We Built It
The skill is engineered according to the official Uliya Skill Development Standard:
- Core Processing Engine: Built with an asynchronous Python 3.11 event daemon running in a sandboxed, unprivileged execution profile on the UGOS Linux layer.
- Deterministic Rules & Parsing: Employs
pydanticschemas for rigorous structural contract enforcement andpdfminer.sixfor deterministic text/table stream extraction. - UGOS CLI Integration: Interacts directly with the platform using mockable UGOS CLI contracts (
ugos-nas fs watch,ugos-nas file tag,ugos-nas audit log) to register audit events directly into the native Skills Hub ledger. - Network Isolation: Runs with strict offline boundaries (
--net=none), enforcing absolute privacy and deterministic repeatability.
Challenges We Faced
- Zero-Cloud Performance Optimization: Designing a low-footprint parsing pipeline capable of running efficiently within NAS thermal and memory envelopes without saturating onboard system resources during parallel ingest operations.
- Robust Error Handling: Ensuring corrupted, truncated, or password-protected PDF streams fail safely without blocking the asynchronous directory listener or causing system service interruptions.
Accomplishments & What We Learned
- Proved that rigorous corporate document auditing and compliance checks can be executed reliably at the edge on consumer/prosumer NAS appliances.
- Designed a lightweight, modular skill specification adhering to the Uliya format, setting up a drop-in architectural baseline ready for physical hardware deployment on the UGREEN AI NAS iDX6011 Pro.
What's Next for Doc-Privacy-Auditor
- Implementing hardware-accelerated local vector embedding leveraging the onboard NPU for instant semantic search.
- Integrating localized webhook alerts to notify designated network administrators via LAN push notifications whenever compliance audits fail.
