Inspiration
Small businesses hold years of personal information across folders, documents, scanned forms and photographs. When a former client exercises their right to access their personal data, finding everything relating to that individual can be time-consuming and error-prone.
My recent work has focused on building trustworthy agentic systems (auditability and security). For this project, my goal was to have a system that can reliably answer a "send me everything you hold on me" request from the files on a small business's NAS. Retrieval is an important capability for any storage system and my aim was to make the search results auditable and honest about its limitations.
What it does
Handles data-subject access and erasure requests against nominated NAS folders. For every file in scope it records what was searched, the method used, what could not be read, matches, uncertain items, third-party data flags, and special-category data flags. Produces a coverage receipt and a triage package. Originals are tagged; copies are placed in dedicated 'release' or 'needs-review' subfolders. No file is deleted by this Skill.
How we built it
I built it using the UGREEN Skill Creation Tool orchestrating specialized NAS capabilities (file-agent, document-agent, photo-agent, knowledge-agent and ug-filemgr). The workflow moves from request intake and file discovery to extraction, matching and human review, before producing the final inventory and coverage report.
Challenges we ran into
The biggest challenge was completeness and accuracy of matches. A name match does not necessarily establish that a document belongs to the requester and a document may contain information about other people. To tackle this, I designed the workflow to distinguish confirmed matches, ambiguous matches, and records requiring human review instead of treating all search results as relevant.
Another challenge was making the search limitations visible. Unreadable scans and inaccessible folders can all create gaps that a regular search might overlook. These limitations are documented in the coverage report.
Accomplishments that we're proud of
I am proud of turning an administrative problem into a structured workflow and designing around human oversight, privacy-conscious processing and non-destructive operations. I believe these are all relevant to the realities of small businesses handling sensitive client information.
What we learned
I enjoyed learning about UGREEN's AI NAS capabilities and exploring how the specialized agents could work together in a defined workflow as this was my first time. I also realized the importance of paying attention to the order in which the agents are called and the state at every point in time (e.g. ensuring I called the file-agent before the knowledge-agent 'collect-evidence'; the initial workflow referencing a folder that was created at a later step). The checker was also quite useful during the process.
What's next for Subject Access Trail
The immediate priority is to validate the complete workflow on UGREEN AI NAS and iterate to improve accuracy and coverage details. Beyond Data Subject Access Requests (DSAR), the underlying workflow can be adapted to other scenarios and usecases (e.g. employee records requests, client offboarding, regulatory audits). The long term vision would be to make reusable records intelligence workflow that helps small businesses understand what information they hold, where it lives, and how it can be responsibly managed.
Built With
- claude