Shiju · AI Signal Archive
Keeping only the ideas worth reading further from podcasts, videos, and articles
Content & Knowledge Product Case / 08
Shiju does not move more content onto another website; it creates a restrained cognitive filter that keeps only ideas worth reading further.
The product organizes podcasts, videos, and articles through filter, compress, extract, and rank, combining a low-density homepage, high-density archive, and immersive excerpts into a reading funnel. AI assists; humans retain final publishing authority.
V1 is live with homepage, archive, detail, and category filtering.
Every item retains its source and at least one key quote.
Homepage features are capped at three, prioritizing quality over volume.
The MVP uses static content data; automated ingestion, full-text search, and membership are not in the current version.
Information Overload Is a Judgment Problem
Users do not lack information; they lack time to judge whether an hour-long podcast is worth it. The product first answers whether to continue, then provides a summary, key ideas, and the original source.
The archive does not republish full texts, preventing another endless feed while respecting original sources.
Three Densities for Three Reading Jobs
The homepage uses space and few selections for discovery; the archive increases density for date and category scanning; the detail view gives visual priority to the strongest idea.
This structure supports a scan in seconds and deeper reading when needed, without forcing every task into one page.
PROCESS
Key Decisions
Curation limits are product rules
No more than three homepage features; absence itself communicates a quality bar.
Source before summary
Every item leads back to its creator; AI compresses but never replaces the source.
Humans own publishing
AI may extract and classify, but a person decides whether an item belongs in the archive.
Role & Collaboration
I led positioning, editorial rules, information architecture, the three-page reading funnel, data contract, and release scope.
AI supports compression and development; topic value, source verification, and publication decisions remain human-owned.
Validation, Outcome & Reflection
The public site is live with working browse, archive, detail, and source-navigation paths.
The current archive is maintained manually; the planned automation pipeline is not presented as operational.
A content product often competes not by covering more, but by clearly deciding what stays out.
