Inspiration
When I return to a research project after a break, I can usually find the old reports. The harder part is remembering which conclusions the evidence supported, which questions were still open, and what I planned to test next. I designed Private Research Memory to make that handoff easier using the documents already stored on a private NAS.
What it does
Private Research Memory is a UGOS Skill designed to turn a small set of authorized research documents into a snapshot with citations, open questions, and one suggested next experiment. When a researcher returns with new evidence, the Skill uses the previous snapshot to show which conclusions changed and which questions still need work.
How I built it
I authored a Markdown Skill package using the official Uliya Skill Kit. The workflow delegates file discovery and snapshot reading to the file expert, evidence extraction to the document expert, and saving an approved snapshot to the knowledge expert. The package passed the official static checker. Document reading, model routing, and snapshot persistence still need to be verified on a NAS.
Challenges and what I learned
The hardest part was deciding how much a new result should change the research record. In a synthetic fruit classification walkthrough, the first report showed 95% accuracy on a random image split. A follow-up showed 78% when the training and test sets contained different fruit IDs. The follow-up gave a better basis for estimating performance on unseen fruit, while overlap in the original split remained an open question. That distinction shaped the snapshot format: each conclusion keeps its source, and unresolved questions carry forward when new evidence leaves them open.
Built With
- markdown
- ugos
- uliya
