[CS] Ozan Gokdemir Candidacy Exam/Jan 12, 2026

[via cs]

This is an announcement of Ozan Gokdemir's Candidacy Exam.
===============================================
Candidate: Ozan Gokdemir

Date: Monday, January 12, 2026

Time: 11 am CST

Location: JCL 298

Title: Enabling Scientific Reasoning at Scale with Retrieval-Augmented Language Models and High-Performance Computing

Abstract:  Scientific discovery is increasingly constrained not by data scarcity, but by the inability to efficiently synthesize and reason over rapidly growing bodies of literature. Large language models (LLMs) offer a promising interface for scientific reasoning, yet their effectiveness depends on scalable access to high-fidelity scientific knowledge and rigorous evaluation of their reasoning behavior. This work explores how high-performance computing (HPC) can be leveraged to enable retrieval-augmented generation (RAG) workflows that support reliable, domain-specific scientific reasoning at scale.
I develop HPC-enabled methodologies for large-scale scientific document ingestion, retrieval, and evaluation that collectively advance the use of LLMs as scientific tools rather than text generators. By coupling adaptive document processing with high-throughput retrieval-augmented inference and automated reasoning benchmarks, this work demonstrates how LLMs can be systematically integrated into scientific workflows to improve knowledge synthesis, question answering, and evaluation across large and heterogeneous scientific corpora. Together, these contributions establish a scalable foundation for AI-driven scientific reasoning on modern supercomputing platforms.

Advisors: Rick Stevens

Committee Members: Rick Stevens, Ian Foster, and Arvind Ramanathan