TSCC as an AI-for-Science Platform: Enabling Research with LLMs, Agentic Tools, and Campus Cyberinfrastructure (Advanced Computing Series)

The TSCC AI webinar explores how the Triton Shared Computing Cluster (TSCC) can serve as an AI-for-science platform by integrating large language models (LLMs) and AI coding agents into the high-performance computing (HPC) research environment. We will introduce several LLM options for TSCC workflows, including models hosted by the National Research Platform (NRP), the UC San Diego TritonAI Developer API, and researcher-managed models deployed directly on TSCC using vLLM and reverse-proxy services. We will demonstrate how agentic AI tools such as Claude Code, OpenCode, and Crush can operate within researchers' TSCC environments to assist with scientific software development, debugging, code optimization, Slurm job submission, and research workflows. We will also discuss the challenges of unmanaged agentic AI usage, including ongoing service costs, research data handling, and the risks of using external AI services without appropriate oversight. By exploring institutionally hosted and self-deployed LLM options, we aim to identify practical approaches that offer researchers greater flexibility, control over AI workflows, and more transparent resource usage. The webinar will share our preliminary benchmarking and evaluation results, highlighting the capabilities and limitations of different LLM deployments and agentic coding tools on TSCC. Finally, we will introduce TSCC's user support, training, and research collaboration opportunities to help researchers explore and adopt AI-assisted scientific computing.