Inspiration
Finding a file is only part of preparing paperwork. For a repair, warranty request or project handover, people also need to know which records belong together and what is missing. Evidence Room grew from that problem.
What it does
Evidence Room is an AI skill for UGREEN AI NAS. The user chooses a folder, describes the case and supplies a checklist, or confirms one suggested by the skill. It finds relevant documents and photographs, puts events in order and prepares a report with references to the original files.
For example, someone dealing with a leak after roof repairs can ask it to gather the quotation, invoice, photographs and notes. The report separates planned work from reported events and flags missing documentation. An invoice does not prove that the work was completed.
How I built it
I used the official UGOS Skill Creation Tool to write a Markdown skill that coordinates registered file, document, photo and knowledge experts.
The workflow requires folder access and permitted model routing to be confirmed before reading. Original files remain unchanged. Saving requires approval of the report and destination, followed by verification of the saved result.
ChatGPT/Codex assisted with planning, writing, simulation and review. No third-party code, libraries or media are bundled.
Challenges I ran into
The main challenge was deciding what a source actually supports. A document may describe an event without confirming it, and two different dates may refer to different stages of the same job.
During a text simulation, a note describing a design was initially treated as though the design file had been found. I added an explicit rule to distinguish a requested file from a description of it.
Accomplishments
The skill and report format are written, and the official checker reports no errors or warnings. Text simulations covered roof records, similar appliance models, project handover and uncertain save outcomes.
The workflow keeps missing information and conflicting accounts visible, and records user corrections separately from the original sources.
What I learned
A useful report needs more than a summary. It should explain why each record is relevant, what it supports and which questions remain unanswered. It also needs to distinguish a failed or incomplete search from a document that was not found in the reviewed material.
What's next for Evidence Room
If selected as a finalist, I will test the skill on the assigned NAS and compare it with ordinary Uliya using the same inputs. I will check source accuracy, missing-item detection, permissions and saving behavior.
The current results are from static checks and text simulations, not NAS execution. Camera recordings, voice notes, folder export and sharing are possible later extensions.
Built With
- ai
- codex
- markdown
- nas
- ugos
- ugreen
- uliya
