Imaging reports are a valuable source of clinical information, but their free-text format poses challenges for clinical operations, AI workflows, quality improvement, and research. This session will explore modern libraries and techniques to efficiently leverage open-source large language models (LLMs) like Llama 3.1 and Mistral for extracting structured data.  

 Participants will gain hands-on insights into prompt engineering, model chaining, agent generation, and performance evaluation. By the end of the session, attendees will be equipped with practical knowledge and example code to integrate LLMs into their radiology data workflows. 

Course Requirements and Downloads 
This is a BYOD Session. Participants who wish to participate in the hands-on portion of this lab will be required to bring their own laptop with required software installed. Requirements will be distributed closer to the meeting.

 Objectives

• Explain the advantages of using open-source models and the requirements for using them in natural language processing tasks. 
• Recognize a collection of libraries available to ease consumption of open-source models, provide language model interchangeability, structured output, chaining, and tool and agent creation. 
• Differentiate prompting techniques and the value and relative performance improvement they bring.