Ian Ingram, an undergraduate computer science student at Southern Oregon University (SOU), recently led a workshop that introduced ACCESS allocations and support alongside an array of large language model (LLM) concepts to a diverse audience of student participants. Ingram’s workshop included hands-on activities featuring various LLM architecture, tokenization, vectorization and key components of Retrieval-Augmented Generation (RAG), such as data ingestion, indexing, querying and optimization.
“The main purpose of the workshop was to empower attendees with practical knowledge of artificial intelligence (AI) – specifically RAG,” Ingram said.
Despite only being an undergraduate, I wanted to share how I used ACCESS allocations like Jetstream2 at Indiana University to leverage my environment and quickly adapt to this fast-moving, disruptive technology – a journey made possible through the strong support from SOU and especially Professor Bernadette Boscoe, who has mentored me throughout my time at the university.
–Ian Ingram, Southern Oregon University computer science student
Boscoe, an assistant professor of computer science at SOU, attributed much of Ingram’s success to his ability to utilize ACCESS resources. “By allowing undergraduates like Ian Ingram to access supercomputers and support, the U.S. National Science Foundation has provided an incredible opportunity,” she said. “In turn, Ian developed this workshop to share with his peers so that they, too, can benefit from ACCESS allocations in their endeavors to explore and expand their work in the AI field and beyond.”
What exactly is RAG? In his workshop, Ingram explained that RAG combines an LLM with a memory retrieval system – effectively grounding AI outputs while integrating semantic search to ensure that responses are informed by relevant, contextual data. He said that context is crucial for accuracy – whether it involves regulatory information or domain-specific inquiries – as AI outputs improve when they’re tailored to the user’s exact circumstances.
“Another critical aspect is safeguarding data – especially when dealing with sensitive or proprietary information – and the solution presented during the workshop involved using open-source frameworks to build offline RAG workflows, ensuring full control over data,” Ingram said. “Tools like Llamaindex, FAISS and Meta’s Llama were discussed as options that can be run on local infrastructure, thus addressing concerns around data privacy while providing the benefits of rapid, AI-powered problem-solving.”
Where does Ingram see himself in the future? He said that he hopes to continue working with ACCESS resources after graduation in a career that allows him to develop organizational strategy and change management in the AI community.
“One of my career goals is to continue to promote inclusive AI resources to help the National Science Foundation empower diverse communities throughout the United States,” he said.
Resource Provider Institution(s): Indiana University (Jetstream2)
Affiliations: Southern Oregon University
Funding Agency: NSF
Grant or Allocation Number(s): CIS240484
The science story featured here was enabled by the U.S. National Science Foundation’s ACCESS program, which is supported by National Science Foundation grants #2138259, #2138286, #2138307, #2137603, and #2138296.