Fix text nodes creation and improve agent handling

[albertgilopez]

This PR addresses several issues and improvements in the multimodal RAG pipeline:

1. Text Nodes Creation:

   • Modified the get_text_nodes function to handle multiple documents and create a node for each page
   • Updated the code where text nodes are created to use the modified function
   • Added checks to print the number of nodes and content of specific nodes for verification

2. Multimodal Query Engine:

   • Updated the MultimodalQueryEngine class to ensure it's properly returning source nodes

3. Agent Setup and Response Handling:

   • Modified the setup for both multimodal and base agents to use ReActAgent.from_tools()
   • Updated the query process to use agent.chat() instead of agent.query()
   • Improved error handling when accessing source nodes in the response

4. Consistency:

   • Ensured consistency between multimodal and base agent setups
   • Standardized the way queries are made and responses are handled for both agents

These changes ensure that:

• We create a text node for each page in the document, matching the number of images
• The query engine properly returns source nodes
• Both multimodal and base agents are set up and queried consistently
• Responses are handled safely, with proper checks for the existence of source nodes

Testing:

• Verified that the number of text nodes matches the number of images
• Tested queries on both multimodal and base agents to ensure they return expected results with source nodes

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB