I would greatly appreciate your guidance or any best practices for implementing human-in-the-loop with LangGraph.

[leeks9653]

We offer a variety of services to our customers.

The functionality I’m trying to implement is as follows:
1. A customer attempts to book a specific service.
2. The system checks whether all required inputs for that service have been provided.
3. If any required data is missing, we ask the customer to provide the missing information.
4. Once all necessary data is collected, we call the service reservation API.

I’m planning to use human-in-the-loop to handle steps 2, 3, and 4.

My approach:
When there is missing information, I trigger an interrupt to inform the customer about what is needed.
Then, once the customer replies, I want to extract any relevant data from their response and update the reservation data object accordingly.

The issue I’m facing:
It seems that every time I use an interrupt and receive a new message from the customer, the LLM gets re-invoked from the beginning. I’d like some advice on the best way to implement this — ideally in a way that avoids unnecessary repeated LLM calls at every interrupt.

`

prev_form_data = {...}

while True: 

    def check_validation():
        ''' 
        When the form data is fully completed: true
        '''
    
    if check_validation():
        # When the form data is fully completed
        pass
    else :
        # When more data is needed
        message = interrupt()

        extract_data = llm.invoke(message)

        prev_form_data = {**prev_form_data, **extract_data}

`