Demo
Summary
The University of Kentucky Center for the Enhancement of Learning and Teaching (CELT) works collaboratively with instructors, staff, and administrators across campus to support engaging, inclusive, and innovative learning environments in which all students and instructors have the opportunity to succeed. CELT curates a wide range of teaching and learning resources. Through this prototype collaboration, they wanted to explore how AI might help website visitors find the information needed more quickly.
Using Retrieval-Augmented Generation (RAG) techniques, CAAI developed a prototype demo that captured CELT’s vetted resources and connected them to users via a familiar conversational AI interface. This allowed users to ask questions in natural language and receive relevant, context-based answers drawn directly from CELT’s materials. While the project was experimental and not intended for broad deployment, it demonstrated the potential of RAG techniques for navigating large, resource-rich websites.
Datasets/Models
The knowledge base of this model was informed by the University of Kentucky’s Center for the Enhancement of Learning and Teaching (CELT) website.
To build the RAG database, content from webpages and linked PDFs was crawled and stored in a searchable vector database. Using LangChain, we developed a pipeline that performed semantic similarity search across the content. With RAG techniques, the system retrieved the most relevant information and injected it into the large language model’s prompt, enabling more accurate and context-rich responses. Unlike traditional keyword search, RAG uses embeddings to capture semantic meaning and context, making it well-suited for navigating large collections of resources.
This prototype was developed using LLM Factory; the base model used was Llama 3 8B.
Access
To learn more about CAAI’s retrieval augmented generation (RAG) and tool calling capabilities, check out the LLM Basics repository on our GitHub for introductory demonstrations of these techniques tailored for use with LLM Factory.
Email ai@uky.edu for more information.
Ownership
This exploratory effort was lead by CAAI wrapped up in September 2024.
RAG and tool calling techniques employed as part of this effort also laid the foundation for future CAAI projects, including BLUE.
Resources
At the foundation of this effort is LLM Factory. The base model used was Llama 3.0 8B.