Integrate langchain-google Search Tool with Grounding Metadata Support

[msaadg]

This PR updates the backend to use the langchain-google search tool, which now supports grounding metadata, replacing the native Google Search API in the web_research node of graph.py. The change ensures compatibility with the new tool's response structure, particularly for handling grounding metadata. Corresponding updates in utils.py were made to process the response_metadata["grounding_metadata"]["grounding_chunks"] field for URL resolution and citation generation. The application has been tested to verify the changes work as expected.

Changes:

• Modified backend/src/agent/graph.py:
  • Updated web_research node to use ChatGoogleGenerativeAI with the langchain-google search tool (GenAITool).
  • Adjusted grounding metadata access to response.response_metadata["grounding_metadata"]["grounding_chunks"].
• Modified backend/src/agent/utils.py:
  • Updated resolve_urls and get_citations to handle the new metadata structure from langchain-google.
• Modified backend/pyproject.toml:
  • Added google-ai-generativelanguage>=0.6.18 to the project dependencies to ensure compatibility with the new search tool.
• Verified functionality by running the application.

Testing:

• Ran the application locally and confirmed that the search functionality and citation generation work correctly with the new tool. Also monitored the logs closely (no issues found).

Related Issues:

• None

Checklist:

• Code changes tested locally
• No breaking changes introduced

[Yoseph-Zuskin]

Nice, the links weren't working for me but this helped! Seems there is a minor bug in backend/src/agent/utils.py lines 71 to 73, should be changed to this to fix:

        modified_text = (
            "".join(modified_text[:end_idx])
            + marker_to_insert
            + "".join(modified_text[end_idx:])
        )