AI-Powered Search for Instagram's Saved Posts Tab
Natural language, vision-powered search for the Instagram saved tab, which currently has none
✳The problem they went after
People save hundreds of Instagram posts over months and then cannot find them again. The saved tab has no search, so retrieval means endless scrolling, opening collection after collection, or giving up. Content saved with real intent goes unused, and the loop between saving something and acting on it never closes.
✳What they actually did
A search bar goes into the saved tab, built for descriptive queries rather than keywords: a travel reel with a blue lake, a coffee recipe from last week. A multimodal pipeline matches the query against embedded representations of saved images, reels, captions and thumbnails. A ranking layer factors in similarity, recency and past interactions, and filters let users narrow further. Measurement covers search activation, query-to-result success, revisit frequency and time in the saved tab.
✳What you can take from it
Worth stealing
- Query-to-result success rate as a quality metric
- Embeddings across images, reels, captions and thumbnails
- Ranking blends similarity, recency and past interactions
- Saved posts reframed as an unfinished intent loop
Read this if
Adding AI search to a surface full of unstructured user-saved content
The transferable bitSaving is treated as an unfinished action, which is what makes this more than a search box: the metric that matters becomes how often people come back to saved posts, not how many searches they run. The write-up also names what structured saved data unlocks next, which is how a small feature earns a larger roadmap.
✳The submission itself
Open the PDF →Reading on a phone? The embedded viewer is cramped — open the PDF in a new tab instead.
