src/csbot/customer_support_chatbot.py

from langgraph.graph import StateGraph, MessagesState
from langchain_core.prompts import ChatPromptTemplate
from langgraph.prebuilt import ToolNode
from src.tools.customer_support_tools import query_knowledge_base, search_for_product_reccommendations, data_protection_check, create_new_customer, place_order, retrieve_existing_customer_orders

import os


class Customer_Support_Bot:
    def __init__(self,llm):
        self.llm = llm
        
        
    def chat_bot(self):
        
        prompt = """#Purpose 

        You are a customer service chatbot for a flower shop company. You can help the customer achieve the goals listed below.

        #Goals

        1. Answer questions the user might have relating to serivces offered
        2. Recommend products to the user based on their preferences
        3. Help the customer check on an existing order, or place a new order
        4. To place and manage orders, you will need a customer profile (with an associated id). If the customer already has a profile, perform a data protection check to retrieve their details. If not, create them a profile.

        #Tone

        Helpful and friendly. Use gen-z emojis to keep things lighthearted. You MUST always include a funny flower related pun in every response."""

        chat_template = ChatPromptTemplate.from_messages(
            [
                ('system', prompt),
                ('placeholder', "{messages}")
            ]
        )

       

        tools = [query_knowledge_base, search_for_product_reccommendations, data_protection_check, create_new_customer, place_order, retrieve_existing_customer_orders]

        llm = self.llm

        llm_with_prompt = chat_template | llm.bind_tools(tools)


        def call_agent(message_state: MessagesState):
            
            response = llm_with_prompt.invoke(message_state)

            return {
                'messages': [response]
            }

        def is_there_tool_calls(state: MessagesState):
            last_message = state['messages'][-1]
            if last_message.tool_calls:
                return 'tool_node'
            else:
                return '__end__'


        graph = StateGraph(MessagesState)

        tool_node = ToolNode(tools)

        graph.add_node('agent', call_agent)
        graph.add_node('tool_node', tool_node)

        graph.add_conditional_edges(
            "agent",
            is_there_tool_calls
        )
        graph.add_edge('tool_node', 'agent')

        graph.set_entry_point('agent')

        app = graph.compile()
        return app