It spawns a bunch of Web Workers to compute embeddings for every post using MediaPipe Text Embedder [2], then compares cosine similarity to the embedding computed for some query text. It displays any posts that are semantically similar enough to the query, based on a similarity range slider the user controls.
It seems to work well on a MacBook Pro M2. On an iPhone 12 it can't keep up. A warning toast will popup if your device is falling behind.
[1] https://docs.bsky.app/blog/jetstream
[2] https://ai.google.dev/edge/mediapipe/solutions/text/text_emb...
