Implement a multi-agent architecture

[almutareb]

We currently use a ReAct agent to handle the main thread, which works fine but can be unpredictable at some time, as it leaves little room to control the behavior.

ToDo:

• Breakdown the current behavior into multiple steps, to gain more control over the intermediate steps.
• Implement it as langgraph

Sources

• Corrective Rag

[almutareb]

Relevant LangGraph examples:

• https://github.com/langchain-ai/langgraph/blob/main/examples/rag/langgraph_adaptive_rag.ipynb