[Feature Request]: Enhance Docling Query Engine: Add PGVector, MongoDB, and Qdrant Support via VectorDBFactory Wrapper

[sitloboi2012]

Is your feature request related to a problem? Please describe.

The current query engine implementation (see docling_query_engine.py) leverages ChromaDB by wrapping its collection into a LlamaIndex ChromaVectorStore for indexing. Meanwhile, the VectorDBFactory class provides a mechanism to create vector database storage with various backends. To improve flexibility and meet our RAG objectives outlined in [Feature Request]: Docling data ingestion to RAG (#688), we need to extend this functionality.

Describe the solution you'd like

1. Review Existing Implementation:

• Examine the current ChromaDB-based query engine implementation.
• Understand how the VectorDBFactory maps the user-selected vector DB to a corresponding LlamaIndex VectorStore.

2. Implement Additional Support:

• Develop wrappers or integration logic for alternative vector databases, specifically PGVector, MongoDB, and Qdrant.
• Ensure that these new wrappers map configuration options correctly to the LlamaIndex-supported VectorStore interfaces.

3. Integration & Testing:

• Integrate the new wrappers with the existing query engine interface.
• Test functionality within the context of the DocumentAgent (Phase 1 DocumentAgent (Phase 1) #438) and ensure compatibility with RAG capabilities.
• Update documentation and examples to reflect the extended support.

Additional context

This enhancement is part of our ongoing effort to make the agent more versatile and not limited to a single vector DB. It builds on recent work (e.g., the merged ChromaDB implementation) and aligns with upcoming changes in retrieve_user_proxy_agent.py to support multiple query engines.

[sitloboi2012]

@AgentGenie @Eric-Shang please help me review this issue and assign it for me 😃 This will be the separate sub-issue build on top and extend the current previous work from in #688 #941