Feedback loop

[guglielmo]

Whenever invoked in the UI, by pressing a command (l), the program wil look into the database, considering:

• notes for fully retrieved (interesting) resources
• resources marked as removed, with respect to the ratings that were proposed
• statistics of relevant resources and categories and sources they come from

An algorithm must be devised, to understand which categories or sources are underperforming. This could be a heuristic algorithm, starting from the statistics.

An AI based algorithm could review the content of the most relevant resources, or the ones that were dubbed as relevant manually, even if ranked low by the algorithm, and suggest new categories. Maybe new sources, as well?

A modal window should warn if incurring costs, probably making an estimate based on the amount of text that will be analyzed, and sent to the LLM.

The system should keep track of all operations on single reosurces:

• early removal (while in resource list view)
• removal while reading the short summary
• fetch of the full content
• generation of a summary or takeaway points
• adding a note
• removal after having fetched the content
• removal after having generated a summary/takeaway points
• removal after having written a note
• exporting it to obsidian
• sharing it

This should be transformed into an interest quantity.

I should then keep all activities related to a resource in the database, and then device an algorithm, that
will check:

• the alignment of each category
• the alignemnt of sources
• possible new categories
• possible new sources
• modifications to the user's profile

The list of possible new sources should be generated beforehand, as a mixture of curated research from gihub awesome's lists, known blogs, or sources, known and respected authors' blogs, ... and an AI analysis of the user's profile.
From the list of possible sources, come the list of possible categories.

Selected sources and categories are a subset of the possible ones.

The users can move their interests in time. That's the idea.

Estimate

Effort: 2-3 weeks (16-20 hours)
Start Date: 2026-03-10
End Date: 2026-03-28

Dependencies

• Requires Add and show notes for Resources #11 (Notes) for user feedback data
• Requires Categories and Sources Statistics #4 (Statistics) for performance metrics
• Requires significant usage data to be meaningful

Related

Part of Learning & Personalization milestone