CGM Comparator
An AI-native research platform for evaluating, comparing, and improving LLM-generated analyses of Continuous Glucose Monitor (CGM) data.
What is CGM Comparator?
CGM Comparator is a research tool that lets clinicians, data scientists, and AI researchers upload CGM time-series datasets, run multiple large language models against them, and compare the resulting analyses side-by-side using objective quality metrics — including functional correctness, efficiency, editability, and semantic similarity.
The platform is designed to answer a key research question: Which AI model best interprets and explains CGM data? By providing a structured evaluation framework, CGM Comparator helps teams make evidence-based decisions when integrating AI into clinical workflows.
Who built this?
Xinxin Yu
University of Waterloo · Graduate Researcher
Xinxin Yu is a graduate researcher at the University of Waterloo with a focus on AI evaluation, clinical decision support, and LLM benchmarking. CGM Comparator was developed as part of research into systematic methods for assessing the quality of AI-generated medical data analyses.
x57yu@uwaterloo.caMission
We believe that rigorous, transparent AI evaluation is essential before AI tools are trusted in clinical settings. Our mission is to give researchers the tools to systematically benchmark AI outputs on medical time-series data, so that the best models rise to the top — and poor ones are identified early.
Technology
CGM Comparator is built on a modern full-stack architecture: a Vue.js frontend for interactive data exploration, a FastAPI Python backend for model orchestration, PostgreSQL for session persistence, and Docker for safe sandboxed code execution. LLM calls are routed through a configurable provider layer supporting GPT-4, Claude, Gemini, DeepSeek, and more.
Medical Disclaimer
CGM Comparator is a research tool and does not provide medical advice. AI-generated analyses on this platform are for research evaluation purposes only and must not be used to make clinical or treatment decisions. Always consult a qualified healthcare professional.