elegant implementation for human-feedback in the loop

[LDXu93]

I'm working on building an agent using RAG and I want to incorporate human feedback as an indicator for the LLM. For instance, if I think the agent doesn't provide the correct answer, it should be able to call the RAG or another tool to try again based on my suggestions. I reviewed the tutorial for human-in-the-loop, but it initiates a new for loop each time a human adds a new message, like this:

for event in app.stream(None, config, stream_mode="values"):  
    event["messages"][-1].pretty_print()  

I'm wondering if there's a way to maintain everything within a single loop, such as:

while True:  
    user_input = input("User: ")  
    clear_output(wait=True)  
    if user_input.lower() in ["quit", "exit", "q"]:  
        print("Goodbye!")  
        break  
    for event in graph.stream({"messages": [("user", user_input)]}):  
        for value in event.values():  
            if isinstance(value["messages"][-1], BaseMessage):  
                print("Assistant:", value["messages"][-1].content)  

[hwchase17]

this is my go-to: add a human node that doesnt do anything, but instead user input just updates state as that node: https://langchain-ai.github.io/langgraph/how-tos/human_in_the_loop/wait-user-input/
can easily maintain place in convo, can easily branch after the input, etc

[LDXu93]

Thank you for your response. I reviewed the tutorial on 'wait-user-input' again to ensure I didn't overlook anything. To clarify, I want to confirm that the human feedback is stored in the user_feedback field within StateSnapshot, and that each update will overwrite the existing user_feedback.

[LDXu93]

I utilized your structured chatbot agent template for tool calling using version v0.1, and incorporating human feedback was straightforward by editing the chat history. However, since our goal is to develop a single-agent system, I find the complexity of the implementation at the tutorial to be confusing.