The codebase indexion is now based on direct chunking and upload to the pgvector database. The RAG built is only taking into account the chunked content.
There is possible optimization and exploration to do, by building a knowledge graph with a graphDB (similar to neo4j) to optimize the RAG on searching the codebase.
- When we index the source of the webpage, even when the agent tries the query the data the right way, it doesn't really work seamlessly.
- A knowledge graph might help figuring out what data could be fetch at which point.
The codebase indexion is now based on direct chunking and upload to the pgvector database. The RAG built is only taking into account the chunked content.
There is possible optimization and exploration to do, by building a knowledge graph with a graphDB (similar to neo4j) to optimize the RAG on searching the codebase.