Back to full roadmap
topiccore
Function / Tool Calling
Give the model function signatures; it picks which to call with what args.
4 hours2 resources1 prereqs
For modern LLMs, this collapses structured output + intent classification + parameter extraction into a single step.
Pattern:
{
"name": "get_weather",
"parameters": {
"city": "Istanbul",
"unit": "celsius"
}
}
The model emits this JSON, you call the real function, then feed the result back. Foundation of agent loops.