Inspiration
Anyone who has worked on a commercial video shoot, an ad campaign, or a documentary knows the dread of pre-delivery wrap-up.
The shoot finishes, and you ingest terabytes of raw 4K footage, b-roll, audio tracks, and still photos straight onto your NAS over a 10GbE network connection. But alongside that creative footage is a messy pile of PDFs and scanned documents: talent release forms, location permits, and music synchronization licenses. Before you can hand the final project over to a client, an ad agency, or a broadcast network, someone has to spend hours answering uncomfortable questions:
- Did every person recognizable in this crowd shot actually sign a release form?
- Does our music license expire next month, or does it cover next year's campaign rollout?
- Are we 100% sure we aren't delivering a file that could trigger a copyright strike or a lawsuit?
Most teams still tackle this manually with spreadsheets and sticky notes because existing AI tools aren't built for this job. You cannot upload unreleased client footage, actor photos, or confidential contracts to cloud AI chatbots—it breaches client NDAs, leaks unannounced creative work, and uploading 20 terabytes over standard office internet would take days or weeks.
I built Provenance Audit Sentinel to solve this problem right where the data already lives. By pairing local AI acceleration with the direct disk access of a UGREEN NAS, I wanted to turn the NAS into a private, on-premise compliance officer that audits creative work in minutes without a single byte ever leaving your local network.
What It Does
Provenance Audit Sentinel runs inside UGOS as a native, private skill. When you point it at a project folder, it quietly conducts a complete audit in five clear steps:
- Takes Inventory: It scans the project folder, catalogs every raw video, audio track, photo, and legal PDF, and calculates a SHA-256 digital fingerprint for each file to guarantee nothing gets lost or corrupted.
- Reads the Paperwork Using local document processing, it reads the agreements in the folder—identifying named actors, permitted distribution regions, and licensing dates.
- Inspects the Media: It runs local facial clustering across the footage and photos to identify key performers, while checking video metadata and audio tracks against your project brief.
- Flags the Gaps: It reconciles the two sides. If a background actor appears on camera without a matching signed release, or if a background music track’s license expires before the campaign launch date, the system immediately flags the asset with an explanation of why.
- Waits for Your Green Light: It never touches, renames, or moves your original files on its own. It presents a clean audit table showing what is cleared and what needs review. Once you click approve, it copies only the cleared assets into a staged delivery folder, re-verifies their checksums, and saves a permanent audit report.
How I Built It
We built this project using the official Uliya Skill Development Standard and the UGOS developer framework. Instead of relying on opaque third-party scripts, the workflow is defined through a clean, declarative orchestration file (SKILL.md) that coordinates native UGOS sub-agents:
file-agent (ug-filesystem, ug-filemgr):Scans folders, indexes files, checks cryptographic integrity, and stages verified copies into the release folder without touching original media.
document-agent (ug-document): Extracts key terms from talent releases and music agreements, pulling out exact names, dates, and clause quotes.
photo-agent (ug-photo): Uses local OCR and on-device face clustering to track who appears in photos and video keyframes.
media-agent (ug-video):Probes video and audio containers to verify frame rates, audio channels, and embedded creation timestamps.
knowledge-agent (ug-knowledge): Generates and saves a clean audit ledger (audit_ledger.csv) and human-readable summary report right in your project workspace.
Because the skill is designed for the UGREEN AI NAS iDX6011 Pro, it takes full advantage of the hardware: The Arc GPU and Intel AI Boost NPU handle face clustering and document reading locally, keeping the machine quiet, cool, and power-efficient. The 16-core Intel Core Ultra 7 processor and direct NVMe storage handle high-speed checksum calculations, ensuring files transfer at full multi-gigabit speeds.
Challenges I Ran Into
Not Overwhelming System Memory on Huge Video Shoots When you point an AI tool at a video project containing hundreds of gigabytes of raw 4K clips, loading all that visual data at once will quickly eat up system RAM. We solved this by designing a rolling batching system: the skill inspects assets in small batches of 16 files at a time, writes down the results, and frees up memory before moving to the next batch. This ensures the NAS runs smoothly with plenty of headroom for normal file-sharing tasks.
Protecting Original Files from Accidental Changes There is nothing worse than an automated script that renames or moves camera files in place, breaking video editing timelines. We instituted a strict non-destructive rule: original footage is treated as read-only. The skill stages copies only into a dedicated release folder, and only after the user explicitly types or clicks their approval.
Handling Ambiguous Legal Text Without Guessing AI models are often too eager to guess or smooth over missing information. In contract law, guessing leads to copyright claims. We built strict fallback rules: if an OCR scan is too blurry to read with high confidence, or if a release form is missing clear territory permissions, the skill refuses to mark it cleared. Instead, it isolates the file into a "Needs Review" queue with the exact page and paragraph cited so a human can make the final call.
What I Learnt & Accomplished
A Clean Static Validation Pass: I verified the entire codebase against the official check-skill.mjs test tool, achieving zero errors and zero warnings while handling all required state transitions cleanly.
Local Compute Is Better for Big Data: Working on this hackathon reinforced that edge AI is a fundamental architectural upgrade for media workflows. Why pay thousands of dollars in cloud storage and bandwidth to send massive files across the internet when your local NAS hardware can process them securely where they already sit?
Safety First: Building deterministic human approval gates into multi-agent workflows is the difference between an interesting tech demo and a dependable tool that professionals can trust with real client work.
What's Next: Finalist Testing on the iDX6011 Pro
When selected for Round 2, I will use the assigned testing window on the physical UGREEN AI NAS iDX6011 Pro to put the skill through three real-world trials:
- Full Production Reel: Testing a complete package of 40 photos, 5 video clips, 4 actor releases, and 2 music licenses to verify that all assets clear and copy over with matching checksums.
- Defect Catching: Running a shoot with an unlicensed actor and an expired audio track to verify that the system catches both defects without moving any unapproved files.
- Edge Case Handling: Throwing corrupted video headers and zero-byte PDFs at the system to ensure it flags the broken files gracefully without crashing.
I look forward to demonstrating the entire workflow running live on the iDX6011 Pro in our finalist demo video!
Disclosures
Platform: UGOS Pro (Linux Debian 12 environment) on UGREEN AI NAS.
Tooling: Official UGREEN uliya-skill-kit (Catalog Version 2.0.0)
Third-Party Code: No external binaries, background daemons, or cloud web APIs are included in the package.
Built With
- ug-document
- ug-filesystem
- ug-knowledge
- ug-photo
- ug-video
