Pixeltable turns your database into a reactive data layer -- embeddings compute automatically on insert. Pair it with Tigris for globally distributed ML apps.

Tigris

A developer built a vibe-based image search app over Nintendo Switch screenshots using Pixeltable as a reactive data layer and Tigris for object storage. Pixeltable's computed columns automatically generate CLIP embeddings on insert, eliminating the need for separate ingestion pipelines, background workers, or manual vector indexing. The post covers the original ambitious plan (video-based combat analytics for an MMO), why it failed due to video compression artifacts, and the pivot to semantic image search. The UI uses Flask and HTMX for simplicity. Key takeaway: Pixeltable's declarative, reactive schema design removes ML data plumbing friction and lets developers focus on the actual application.

How I learned to love reactive data models with Pixeltable

From grand vision to pragmatic haircut ​