Skip to content
Lecture Revision Lab

Step 2 · Retrieve

Find the passages that answer a question

The same retriever as the 2024 prototype: all-MiniLM-L6-v2 embeddings, cosine similarity, top k. Embeddings are now computed once and cached instead of on every query.

Retrieval explorer

Pick a sample question or type your own.

By default every passage is scored by cosine similarity to your query, exactly like the 2024 retriever, and the top k come back with matching words highlighted. BM25 and a hybrid are one click away. No API key needed.