Agentic AI for high-frequency data quality monitoring

This project focuses on advancing the automated quality assessment of high-frequency physiological recordings through OUCRU Vital Agent, an explainable agentic AI system for ECG and PPG signal quality control. The primary objective is to replace slow and inconsistent manual review with a tool-driven pipeline that identifies low-quality signal regions, preserves the supporting evidence, and reduces the time required to prepare quality-control reports for clinical research.



Core activities include implementing a deterministic signal-quality pipeline in which approved signal-processing tools generate SQI results and the language model is restricted to explaining stored evidence. The data model is organised around an Organization–Study–Patient–Recording–Segment hierarchy, enabling every result and report to remain linked to its research context. A staged review workflow is developed comprising SQI scoring, AI-assisted review, human review and approval, and final report generation with FHIR export.

System development utilises FastAPI, Next.js, PostgreSQL and Docker, with a self-hosted Qwen3-8B runtime deployed on Google Cloud. Signal processing is performed using the vital_sqi, WFDB and NeuroKit2 libraries.

The contribution is a traceable, auditable and portable quality-control platform that converts hours of expert review into minutes of verified decision-making, establishing a reusable standard for evidence-grounded AI in clinical research data pipelines.