Function-Calling and Data Extraction with LLMs

This course will teach you two critical skills for building applications with LLMs: function-calling and structured data extraction.

Function-calling allows you to extend LLMs with custom capabilities by enabling them to form calls to external functions based on natural language instructions. Structured data extraction enables LLMs to pull usable information from unstructured text. 

You’ll work with NexusRavenV2-13B, an open source model fine-tuned for function-calling and data extraction. The model, available on Hugging Face, has outperformed GPT-4 in some function-calling tasks, and has 13 billion parameters so it can be hosted locally. 

What you’ll explore: 

• Learn how you can use function-calling in detail: form prompts with function definitions, and use an LLM response to call those functions.
• Use an LLM with multiple function calls, including parallel and nested function calls. This allows you to create complex agent workflows where an LLM plans and executes a series of function calls to achieve a goal.
• Use OpenAPI specifications to build function calls that can access web services.
• Use function-calling to extract structured data from a natural language input. 
• Build an application that takes customer service transcripts, builds SQL calls, and stores results in a database with commands generated by the LLM. 

The skills you’ll learn in this course will allow you to build advanced AI agents and assistants that can process and analyze customer feedback, automate data entry and content management workflows, enhance search and recommendation systems with structured data, and many other real-world applications.

This course is designed for anyone interested in using function-calling capabilities with large language models to interface with external tools and extract structured data. Familiarity with LLMs and basic Python knowledge are recommended.