The problem
People often remember a photo only in pieces: a colour, a place, who was there. They can describe it, but Google Photos search often can't match that description to the photo. Search works on dates and faces, but not on descriptions.
What I found
- I collected 705 public complaints (Play Store, App Store, Reddit) and used an LLM to pick out 84 about retrieval.
- 59 of those 84 break at one stage: search can't read the clue the person gives.
- I interviewed 6 people. 3 of 6 described the photo clearly and still failed to find it.
What I built
An AI search prototype. Every photo has a plain-language caption plus its date, place and people. The typed memory and the captions go to an LLM, which returns the photos that fit, each with a 0–100 confidence score. If nothing is confident enough, it shows "no confident match" instead of guessing.
Results
I tested the prototype with 3 users from the same group. All 3 rated it above their own photo app (for example 4–4.5 out of 5 against 1 out of 5 for one tester). It never invented a match on a photo that wasn't there.
Links
Try the prototype: https://gp-retrieval-mvp.netlify.app/
Explore the complaint data: https://retrieval-radar.netlify.app/