👐 Run Vector Queries

Let's perform vector queries against our vector search index.

First, we'll vectorize the user's query (text or image) using the same CLIP model we used for the book covers.

Then we'll use these query vectors to search for semantically similar book covers in our vector search index.

CODE_BLOCK_8

Answer

get_embedding(user_query, mode)

CODE_BLOCK_9

Answer

[
    {
        "$vectorSearch": {
            "index": ATLAS_VECTOR_SEARCH_INDEX_NAME,
            "queryVector": query_embedding,
            "path": "embedding",
            "numCandidates": 50,
            "filter": filter,
            "limit": 5,
        }
    },
    {"$project": {"_id": 0, "title": 1, "score": {"$meta": "vectorSearchScore"}}},
]

CODE_BLOCK_10

Answer

collection.aggregate(pipeline)