Alexander Ponomarev
Research Mentor: Fadhl Alakwaa
Mentor Department: Internal Medicine Department, Medicine
Author(s): Alexander Ponomarev, Fadhl Alakwaa
Session: Session 5 (2:00 PM – 2:50 PM)
Presentation Type: Poster 65
Abstract
Query2Plot is an AI-powered assistant that enables Michigan Medicine clinicians to safely and efficiently analyze biomedical datasets—including structured clinical research tables and high-dimensional molecular data—within a fully HIPAA-compliant environment. Clinicians face major barriers to using modern AI tools because most commercial or cloud-hosted analytics platforms cannot be applied to sensitive clinical data due to privacy, security, and regulatory constraints. As a result, analyses often rely on manual workflows or fragmented pipelines that slow discovery and limit the timely generation of clinically relevant insights. Query2Plot addresses this gap by hosting all data ingestion, processing, and model execution entirely within the University of Michigan’s secure high-performance computing infrastructure, ensuring that electronic protected health information (ePHI) remains within institutional boundaries. The system supports exploratory analysis, statistical summarization, and reproducible workflows through a clinician-friendly interface without any data leaving the secure environment. At its core, Query2Plot uses an agent-based architecture built with CrewAI (v1.7.2) and leverages GPT-4o-mini for task orchestration, code generation, and interpretation. The assistant modularly coordinates schema inference, preprocessing, analysis planning, sandboxed code execution, and results reporting under strict access controls and execution guardrails. This design returns interpretable summaries and publication-ready figures suitable for clinical and translational research workflows while maintaining the security standards required in clinical settings. A public (non-PHI) version of the platform is available at: https://huggingface.co/spaces/russkiyximik/query2plot
