SafeVault AI - Local Medical & Document Assistant

Inspiration

In an era where convenience often comes at the cost of digital privacy, storing sensitive personal documents such as medical records, tax forms, and national identification cards on public cloud services exposes users to severe data security risks. SafeVault AI was inspired by the need for absolute data sovereignty, leveraging local hardware to manage personal documents securely without relying on third-party cloud servers.

What it does

SafeVault AI is an automated, zero-cloud NAS Skill designed specifically for UGREEN AI NAS (UGOS). It acts as an autonomous background agent that:

• Monitors local drop directories for incoming document scans.
• Performs on-device OCR and visual parsing.
• Automatically categorizes, renames, and moves files into a structured local directory.
• Generates privacy-protected, auto-redacted copies for safe external sharing.
• Updates an encrypted local metadata index (emergency_summary.json) for instant emergency retrieval.

How we built it

The project is built using a lightweight Python-based pipeline integrated into the UGOS environment:

• OCR & Parsing: Utilizes open-source local engines (Tesseract / PaddleOCR) and lightweight regex parsing for PII detection.
• NAS Orchestration: Leverages UGOS CLI capabilities (ugos-cli file move, ugos-cli metadata set) to handle file permissions and directory management locally.
• Standard Skill Structure: Packaged following the Uliya Skill Development Standard via a clean SKILL.md agent specification file.

Challenges we ran into

• Hardware Optimization: Ensuring that local OCR and regex scanning execute smoothly on embedded NAS hardware without bottlenecking system resources.
• Accurate PII Masking: Designing a reliable bounding-box masking mechanism via OpenCV to automatically obscure sensitive numbers (such as ID/NIK numbers) on generated shareable copies.

Accomplishments that we're proud of

• Successfully designing a 100% offline, privacy-first document workflow architecture tailored specifically for the UGREEN AI NAS ecosystem.
• Creating a dual-copy safety mechanism that automatically separates restricted master files from privacy-safe shareable documents.

What we learned

We gained deep insights into edge-AI processing, local agent orchestration within closed storage systems, and the critical importance of implementing privacy-first architectures directly at the storage layer.

What's next for SafeVault AI - Local Medical & Document Assistant

• Implementing advanced multilingual OCR support for multi-regional documents.
• Integrating fine-tuned lightweight local LLM embeddings for smarter semantic document search directly on the NAS hardware.
• Expanding the automated categorization rules for a wider variety of financial and legal documents.

Built With

• ai
• automation
• computing
• document
• edge
• local
• management
• nas
• ocr
• open
• privacy
• python
• source
• ugos
• ugreen
• zero-cloud