{ "cells": [ { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "from langchain.chat_models import ChatOpenAI\n", "from langchain.memory import ConversationBufferMemory\n", "from langchain.prompts import PromptTemplate\n", "from langchain.chains import LLMChain\n", "from langchain.utilities import SQLDatabase\n", "from sqlalchemy import create_engine # Import create_engine\n", "\n", "# --- Initialize Core Components ---\n", "\n", "# 1. Dialogue Context (Memory)\n", "memory = ConversationBufferMemory()\n", "\n", "# 2. LLM (for routing, service selection, state tracking, and response generation)\n", "llm = ChatOpenAI(temperature=0, model_name=\"gpt-3.5-turbo\") # Or another suitable model\n", "\n", "# 3. Database (using SQLite in-memory for demonstration)\n", "engine = create_engine(\"sqlite:///:memory:\") # Create an in-memory SQLite engine\n", "db = SQLDatabase(engine) # Pass the engine to SQLDatabase\n", "\n", "# --- Define Prompts ---\n", "\n", "# Router Prompt\n", "router_template = \"\"\"\n", "You are a helpful assistant that classifies user input into two categories:\n", "\n", "1. open-domain: General conversation, chit-chat, or questions not related to a specific task.\n", "2. task-oriented: The user wants to perform a specific action or get information related to a predefined service.\n", "\n", "Based on the dialogue history, classify the latest user input:\n", "\n", "{chat_history}\n", "\n", "User: {user_input}\n", "\n", "Classification:\n", "\"\"\"\n", "router_prompt = PromptTemplate(\n", " input_variables=[\"chat_history\", \"user_input\"], template=router_template\n", ")\n", "\n", "# Service Selection Prompt\n", "service_selection_template = \"\"\"\n", "You are a helpful assistant that classifies user input into one of the following predefined services:\n", "\n", "Services:\n", "- book_flight: For booking flight tickets.\n", "- check_order_status: For checking the status of an order.\n", "- find_restaurants: For finding restaurants based on criteria.\n", "\n", "Based on the dialogue history, which service best matches the user's intent?\n", "\n", "{chat_history}\n", "\n", "User: {user_input}\n", "\n", "Selected Service:\n", "\"\"\"\n", "service_selection_prompt = PromptTemplate(\n", " input_variables=[\"chat_history\", \"user_input\"],\n", " template=service_selection_template,\n", ")\n", "\n", "# Dialogue State Tracking Prompt\n", "state_tracking_template = \"\"\"\n", "You are a helpful assistant that extracts information from user input to fill in the slots for a specific service.\n", "\n", "Service: {service}\n", "Slots: {slots}\n", "\n", "Based on the dialogue history, extract the values for each slot from the conversation. \n", "Return the output in JSON format. If a slot is not filled, use null as the value.\n", "\n", "{chat_history}\n", "\n", "User: {user_input}\n", "\n", "Extracted Information (JSON):\n", "\"\"\"\n", "state_tracking_prompt = PromptTemplate(\n", " input_variables=[\"service\", \"slots\", \"chat_history\", \"user_input\"],\n", " template=state_tracking_template,\n", ")\n", "\n", "# Response Generation Prompt\n", "response_generation_template = \"\"\"\n", "You are a helpful assistant that generates natural language responses to the user.\n", "\n", "Dialogue History:\n", "{chat_history}\n", "\n", "User: {user_input}\n", "\n", "{slot_info}\n", "\n", "{db_results}\n", "\n", "Response:\n", "\"\"\"\n", "response_generation_prompt = PromptTemplate(\n", " input_variables=[\"chat_history\", \"user_input\", \"slot_info\", \"db_results\"],\n", " template=response_generation_template,\n", ")\n", "\n", "# --- Define Chains ---\n", "\n", "router_chain = LLMChain(llm=llm, prompt=router_prompt, output_key=\"classification\")\n", "service_selection_chain = LLMChain(\n", " llm=llm, prompt=service_selection_prompt, output_key=\"service\"\n", ")\n", "state_tracking_chain = LLMChain(\n", " llm=llm, prompt=state_tracking_prompt, output_key=\"slot_json\"\n", ")\n", "response_generation_chain = LLMChain(\n", " llm=llm, prompt=response_generation_prompt, output_key=\"response\"\n", ")\n", "\n", "# --- Define Service Slots ---\n", "# (In a real application, this would likely be loaded from a configuration file or database)\n", "service_slots = {\n", " \"book_flight\": [\"destination\", \"departure_date\", \"num_passengers\"],\n", " \"check_order_status\": [\"order_id\"],\n", " \"find_restaurants\": [\"cuisine\", \"location\", \"price_range\"],\n", "}\n", "\n", "# --- Main Dialogue Loop ---\n", "\n", "def process_user_input(user_input):\n", " # 1. Add user input to memory\n", " memory.chat_memory.add_user_message(user_input)\n", "\n", " # 2. Route the input\n", " router_output = router_chain(\n", " {\"chat_history\": memory.load_memory_variables({}), \"user_input\": user_input}\n", " )\n", " classification = router_output[\"classification\"].strip()\n", "\n", " print(f\"Router Classification: {classification}\")\n", "\n", " if classification == \"open-domain\":\n", " # 3. Handle open-domain conversation\n", " llm_response = llm(memory.load_memory_variables({})[\"history\"])\n", " response = llm_response.content\n", " else:\n", " # 4. Select the service\n", " service_output = service_selection_chain(\n", " {\"chat_history\": memory.load_memory_variables({}), \"user_input\": user_input}\n", " )\n", " service = service_output[\"service\"].strip()\n", "\n", " print(f\"Selected Service: {service}\")\n", "\n", " if service not in service_slots:\n", " response = \"I'm sorry, I cannot understand that service request yet. We currently support booking flights, checking order status and finding restaurants only.\"\n", " else:\n", " # 5. Track the dialogue state (slot filling)\n", " slots = service_slots[service]\n", " state_output = state_tracking_chain(\n", " {\n", " \"service\": service,\n", " \"slots\": \", \".join(slots),\n", " \"chat_history\": memory.load_memory_variables({}),\n", " \"user_input\": user_input,\n", " }\n", " )\n", " slot_json_str = state_output[\"slot_json\"].strip()\n", "\n", " print(f\"Slot Filling Output (JSON): {slot_json_str}\")\n", "\n", " try:\n", " import json\n", " slot_values = json.loads(slot_json_str)\n", " except json.JSONDecodeError:\n", " slot_values = {} # Handle cases where JSON decoding fails\n", " response = \"I'm sorry, there seems to be a problem understanding your request details.\"\n", "\n", " # (Optional) 6. Database interaction (based on service and filled slots)\n", " db_results = \"\" # Initialize db_results as an empty string\n", " if service == \"check_order_status\" and \"order_id\" in slot_values:\n", " try:\n", " order_id = slot_values[\"order_id\"]\n", " # Basic query without table information\n", " db_results = db.run(f\"SELECT * FROM orders WHERE order_id = '{order_id}'\")\n", " db_results = f\"Database Results: {db_results}\"\n", " except Exception as e:\n", " print(f\"Error during database query: {e}\")\n", " db_results = \"\"\n", "\n", " # 7. Generate the response\n", " response_output = response_generation_chain(\n", " {\n", " \"chat_history\": memory.load_memory_variables({}),\n", " \"user_input\": user_input,\n", " \"slot_info\": f\"Slots: {slot_json_str}\",\n", " \"db_results\": db_results,\n", " }\n", " )\n", " response = response_output[\"response\"]\n", "\n", " # 8. Add the system response to memory\n", " memory.chat_memory.add_ai_message(response)\n", "\n", " return response\n", "\n", "# --- Example Usage ---\n", "\n", "while True:\n", " user_input = input(\"You: \")\n", " if user_input.lower() == \"exit\":\n", " break\n", " response = process_user_input(user_input)\n", " print(f\"AI: {response}\")" ] } ], "metadata": { "kernelspec": { "display_name": "crawl_data", "language": "python", "name": "python3" }, "language_info": { "name": "python", "version": "3.10.13" } }, "nbformat": 4, "nbformat_minor": 2 }