Function Calling

Function calling was the first widely adopted standard for LLM-tool integration. Most frameworks and providers now use compatible schemas.

Function calling is placing an order at a deli counter — you specify the item (function) and modifications (arguments); the kitchen (runtime) prepares it.

Scenario

GPT-4o receives a weather tool definition and responds with tool_calls: [{function: {name: 'get_weather', arguments: '{"city": "Berlin"}'}}].

Solution

In Tool Calling & Function Calling, apply Function Calling to this scenario: GPT-4o receives a weather tool definition and responds with tool_calls: [{function: {name: 'get_weather', arguments: '{"city": "Berlin"}'}}]. Identify the inputs, run the technique, validate the output, and note one thing you would monitor in production.

Do this before moving to the next module — reading alone is not enough.

Open the Code Walkthrough below and run it locally. Change one parameter related to Function Calling (e.g. model, temperature, top_k, or tool name), observe the difference in output, and write 2–3 sentences explaining what changed.

Highlighted lines show where Function Calling happens in the code.

1response = client.chat.completions.create(  # call the API2    model="gpt-4o-mini",3    messages=[{"role": "user", "content": "Weather in Berlin?"}],4    tools=[{"type": "function", "function": WEATHER_FN}],  # pass tool schemas so the LLM can call functions5    tool_choice="auto",6)

• Treating Function Calling as a black box without evaluation
• Ignoring cost and latency in production
• Skipping error handling for function calling

Quick recap — the most important points from this module.