Sreyes Venkatesh

Ai Developer in Cardio-thoracic Radiology @University Hospitals

Cleveland, OH, US

I am building a Virtual Resident (VR) AI Agent for automated chest X-ray (CXR) interpretation, leveraging emerging multimodal large language models (MLLMs). This work explores how small, resource-efficient generative AI models can support radiology reporting in real-world hospital environments. The work demonstrates both the promise and current limitations of small multimodal AI models for radiology. By stress-testing generative AI “out-of-the-box,” we aim to establish the requirements for safe, scalable clinical adoption of AI-driven report generation.Key focus areas:Evaluating clinical readiness – benchmarking MLLMs on open-access datasets (IU-CXR and ReXGradient) to measure factual accuracy, lexical alignment, and error rates compared to radiologist reports.Uncovering generalizability gaps – analyzing how performance degrades when moving from controlled academic datasets to heterogeneous, real-world CXR distributions that better reflect clinical practice.Improving model integration – identifying limitations of benchmark-centric evaluation and highlighting the need for fine-tuning, diverse dataset validation, and clinically aligned scoring metrics before deployment.Advancing radiology workflows – contributing to the vision of “Virtual Residents” that can reduce reporting delays, support radiologists in high-volume environments, and expand access to diagnostic tools globally.Accepted at RSNA 2025:\"Out-of-the-Box but Not Out-of-the-Woods: Evaluating Performance of Small Multimodal Generative AI for Chest X-Ray Interpretation\" Sreyes Venkatesh, Syed Muhammad Awais Bukhari, MD, Charit R. Tippareddy, MD, Komal M. Awan, MD, Amit Gupta, MD

University of California, Santa Cruz

B.S, Electrical Engineering and Computer Science

University of Southern California

MASTER OF SCIENCE