Implementing LLM Function Calling with OpenAI Tool Integration

steps

1. Set up your Python project.

   • Ensure the openai library is installed.
   • Prepare your Python files, including products.py to simulate a product database.
   bashCopy

   pip install openai

2. Define a dictionary of product data.
   This will serve as the datasource for your product queries. Save this as products.py.

   pythonCopy

   products = {
       "Ethiopian_Yirgacheffe": {
           "name": "Ethiopian Yirgacheffe",
           "origin": "Ethiopia",
           "flavor_profile": "Bright and citrusy with floral aroma",
           "price": 18.99,
           "certifications": ["Fair Trade", "Organic"]
       }
   }

3. Create a function to fetch product information.

   Define a Python function using typing annotations that extracts requested product details.

   pythonCopy

   from typing import Dict
   
   def get_product_info(product_id: str) -> Dict:
       """
       Fetch details for the specified product.
       """
       if product_id not in products:
           raise ValueError("Product not found.")
       return products[product_id]

4. Describe the function using a JSON schema for the LLM.

   The schema serves as a "contract" that explains the function and its parameters to the language model.

   pythonCopy

   import json
   
   tools = [
       {
           "type": "function",
           "name": "get_product_info",
           "description": "Fetch detailed information about Global Java Roasters products.",
           "parameters": {
               "type": "object",
               "properties": {
                   "product_id": {
                       "type": "string",
                       "description": "The ID of the product to retrieve information for."
                   }
               },
               "required": ["product_id"]
           }
       }
   ]

5. Connect to the OpenAI API and set up the LLM message flow.

   Create the flow that routes user messages, handles tool invocation, and returns structured results.

   pythonCopy

   import openai
   import os
   
   openai.api_key = os.getenv("OPENAI_API_KEY")
   
   messages = [
       {"role": "system", "content": "You are a product assistant for Global Java Roasters."},
       {"role": "user", "content": "Can you tell me about Ethiopian Yirgacheffe?"}
   ]
   
   response = openai.ChatCompletion.create(
       model="gpt-4-0613",  # The name of the OpenAI model.
       messages=messages,
       functions=tools,
       function_call="auto",  # Let the model decide whether to call a function.
   )
   
   tool_request = response["choices"][0]["message"]["function_call"]
   if tool_request:
       function_name = tool_request["name"]
       arguments = json.loads(tool_request["arguments"])
   
       # Call the function
       if function_name == "get_product_info":
           result = get_product_info(**arguments)
   
       # Append the tool response to messages
       messages.append({"role": "tool", "name": function_name, "content": json.dumps(result)})
   
       # Continue the conversation
       final_response = openai.ChatCompletion.create(
           model="gpt-4-0613",
           messages=messages,
       )
   
       print(final_response["choices"][0]["message"]["content"])