AIM 2026 will use a new AI-based sorting process (SMART - the Session Matching and Automated Recommendation Tool) to conduct the initial categorization of presentations and to monitor the process of session organization. 

AIM Program Vice Chair Joe Dvorak will cover:

• This new sorting process and how it works
• The technologies that power the processes 
• Results from AIM 2025 where parts of it were trialed and from the ongoing organization of AIM 2026.

This process does not seek to replace the human organizers. Instead, we replace the name guessing game that committees do at AIM as well as the guessing game that presenters do when trying to select the most appropriate session title for their work. This process uses LLM embedding models and traditional machine learning clustering models to group the presentations in suggested initial sessions. Generative LLMs are then used to summarize and describe these initial session groups so that human organizers can be assigned to assist in the process. Additionally, human organizers are provided a semantic search tool that they can use to find additional on topic presentations if they need to add to their sessions.

A major theme of this work was keeping humans in the loop for what they do best while letting AI systems handle issues that do well such as analyzing the topics of over 1000 presentations quickly. Attendees to this session will learn about building systems using LLMs, embedding models and generative AI to automate difficult tasks. They will also learn about how LLMs enable rapid software development and considerations for keeping humans in the loop.

Joe Dvorak is an Associate Professor and the Director of Undergraduate Studies in the Biosystems and Agricultural Engineering Department at the University of Kentucky. His research focuses on automation through agricultural artificial intelligence.