Please join us on Friday, March 20, 2026, from 12:30 to 1:30 p.m. for the in-person ISP Forum in SENSQ 5317.  We will be featuring two ISP students,  Jonah Belback and Manurag Khullar. Lunch and refreshments will be provided, starting at noon.

Jonah Belback Presentation Information

Title: Large Language Models and Domain Knowledge for Enhanced Multimodal Learning in Predicting Breast Cancer Recurrence Risk

Abstract: Despite the effectiveness of adjuvant endocrine therapies such as Tamoxifen and Aromatase Inhibitors, their long-term use and association with significant side effects causes poor adherence for patients unsure on their individual benefit. In order to address this issue compromising patient recovery, an early response biomarker/method measuring the clinical efficacy of Tamoxifen and Aromatase is needed. In this
work, we proposed a novel method of autonomously integrating domain-specific knowledge into a Transformer-based model to facilitate multimodal learning by leveraging large language models (LLMs) and medical domain knowledge represented by textual data through semantic embeddings and knowledge graphs for enhanced recurrence risk prediction in estrogen positive breast cancer patients. We evaluated the performance of our proposed method on a dataset of 679 patients. Our proposed models achieved superior AUC scores when compared to baseline models, in distinguishing recurrence-free patients from those with new primary cancers. This study showed the promise of our novel technical methods and proved the concept that LLMs and domain knowledge can be a useful source of intelligence to enhance AI models, which is a significant advance for multimodal AI learning. With our enhanced prediction models, it can minimize unnecessary harms from Tamoxifen or Aromatase treatments for patients unlikely to benefit from them and increase patients’ adherence for those more likely to benefit.

Bio: First year master’s student at the University of Pittsburgh’s Intelligent Systems Program with a bachelor's in computer engineering. Focused on the field of Biomedical Informatics, exploring new methodology of incorporating textual domain knowledge into ML imaging architecture.

Manurag Khullar Presentation Information

Title: Script Gap: Evaluating LLM Triage on Indian Languages in Native vs Roman Scripts in a Real-World Setting

Abstract: Large Language Models (LLMs) are increasingly deployed in high-stakes clinical applications in India. In many such settings, speakers of Indian languages frequently communicate using romanized text rather than native scripts, yet existing research rarely evaluates this orthographic variation using real-world data. We investigate how romanization impacts the reliability of LLMs in a critical domain: maternal and newborn healthcare triage. We benchmark leading LLMs on a real-world dataset of user-generated queries spanning five Indian languages and Nepali. Our results reveal consistent degradation in performance for romanized messages, with F1 scores trailing those of native scripts by 5-12 points. At our partner maternal health organization in India, this gap could cause nearly 2 million excess errors in triage. Crucially, this performance gap by scripts is not due to a failure in clinical reasoning. We demonstrate that LLMs often correctly infer the semantic intent of romanized queries. Nevertheless, their final classification outputs remain brittle in the presence of orthographic noise in romanized inputs. Our findings highlight a critical safety blind spot in LLM-based health systems: models that appear to understand romanized input may still fail to act on it reliably.

Bio: Manurag is a first-year PhD student in the Intelligent Systems Program. His research interests include multilingual NLP,  responsible AI, and AI for social good.  Previously, he was a Data Science for Good Fellow at the University of Washington, Seattle Campus, and worked as a technology policy analyst with NITI Aayog, the Government of India’s apex think tank. He holds a master’s degree from the University of Pennsylvania and completed a thesis on explainable AI under Professor Li Shen.