Awesome - I knew about vectorising/embeddings for semantic search, but I hadn't thought of using the search results as a prompt prefix - clever!
Use SebtenceTransformers in python to write to the database (PineconeDB) and then do the same for queries. Use the results as context.