FNU Shilpika
Postdoctoral Appointee
Biography
Shilpika is a Postdoctoral Appointee at the Argonne Leadership Computing Facility. Her research focuses on data analysis and visualization of high-performance computing systems, including interpretation of AI for science to enable informed decision-making in AI workflows. Shilpika obtained an MS in Computer Science from Loyola University Chicago and a Ph.D. in Computer Science from the University of California Davis.
Research Summary
Shilpika develops Explainable AI and high-performance computing tools that help scientists and system operators monitor and manage large computing systems and guide decision-making for AI-driven science workflows.
Part of her work focuses on analyzing system logs to detect patterns that might indicate problems or inefficiencies. She builds methods that can analyze these logs at a very large scale and high fidelity while still producing results that people can understand and interpret.
Her research studies system behavior at multiple levels and time scales, which helps improve tasks such as scheduling jobs on the machine, evaluating performance, and detecting unusual activity or failures. A key focus of her work is ensuring results are explainable and reproducible so others can understand and verify the findings.
She also develops real-time data pipelines that collect information from sensors and system telemetry so operators can quickly see how a supercomputer is performing. One major project is a digital twin of the Aurora supercomputer — a virtual model that mirrors the real system and helps diagnose issues, predict performance, and test changes safely.
Shilpika builds reusable data pipelines for scientific research. For example, she has developed tools that process large biological data sets to study protein–protein interactions, connecting steps such as data collection, feature extraction, and machine-learning training.
She has also developed an explainable AI method that guides the prompt engineering process of large language models. The method can guide the training of these large language models, leading to more trustworthy and safe AI.
The tools she builds are designed to run in real operational environments, helping teams monitor, diagnose, and make decisions relating to exascale supercomputers and AI for Science.
Selected Awards
- 2025 Outstanding Postdoctoral Performance Award, Argonne National Laboratory
- 2nd Runner-up 2024 Argonne Postdoc Research Slam, Argonne National Laboratory
- 2016 Turing Highest Achievement Award, Loyola University Chicago