Converting existing rag to graph-db without original documents

[Prateek219]

Any views on below idea.
Idea is about converting an existing Pinecone vector database (which has lost track of the original documents) into a Graph RAG system in Neo4j. I can only fetch by IDs and then use the chunks to build a knowledge graph.I also want to explore the Pinecone database by randomly sampling vectors, getting their top-k similar vectors, processing them, and marking them as processed. We then store the extracted knowledge in Neo4j.

Motivation: I have a large Pinecone vector database but the original documents are lost. I want to build a Graph RAG system to capture relationships between entities and enable more sophisticated querying.

Approach:

Connect to Pinecone and Neo4j.

Sample random vectors from Pinecone and fetch their top-k similar vectors.

Extract entities and relationships from the text chunks.

Store the chunks, entities, and relationships in Neo4j.

Mark the processed chunks to avoid reprocessing.