I’m currently using the following format:
- Store all my data as a blob on cloud storage
- Store embeddings and id of the blob on Pinecone
When someone searchs:
- I find the most relevant ids
- Locate the blob on cloud storage
- Filter for relevant info (based on vector search)
- Return to client
Does this flow make sense? The other option is to store in “metadata” but I tried this and hit a limit on the max size. I wondered if my current approach introduces a lot of latency.