How Search Works

Asclevor converts your query into an embedding and searches the vector index for semantically similar cases.

1. Your natural-language query is embedded into a vector using our embedding model.
2. The vector is compared against precomputed case embeddings using cosine similarity.
3. The search runs over an HNSW index, which returns approximate nearest neighbors at low latency.
4. Results are returned ranked by similarity score, highest first.

Because matching happens in embedding space, results are semantic: a query like "hypotonia with developmental delay" can surface cases that use different phrasing, such as "decreased muscle tone and global developmental delay".