Hey everyone, I m one of the makers of Vecstore. We launched here about a year ago, and since then we ve rebuilt most of the stack. Sharing what we learned and what changed:
New model: understands both photos and text, with noticeably better accuracy
Rust embedding service: an image gets embedded in about 12ms
Moved everything into Postgres: queries take about 3ms on 10M records
Hybrid ranking: semantic search alone kept missing exact names and SKUs, so we blended in keyword matching
It s built for marketplaces, ecommerce, and any app with lots of photos. It also handles NSFW detection and duplicate image detection.
You can try the demo without signing up: https://vecstore.app/demo
Hi Product Hunt, we launched Vecstore here about a year ago. Since then we've rebuilt most of it.
What's new:
- A new model that understands photos and text with higher accuracy
- The embedding service rewritten in Rust, with an image embedded in about 12ms
- Everything moved into Postgres, and queries now take about 3ms on 10M records
- Hybrid ranking, so exact names and SKUs still match
It works for marketplaces, ecommerce and any app with lots of photos. It also includes NSFW detection and duplicate image detection.
The demo needs no signup: https://vecstore.app/demo
We'd love feedback on search quality.