Projects
Open source tools, research systems, and clinical AI infrastructure built across academia, NIH-funded grants, and independent work.
AI agents designed for real clinical workflows. Covers ambient documentation (auto-generating SOAP notes from clinician-patient conversations), smart order suggestions, and prior authorization automation — reducing administrative burden on healthcare providers.
Deny-by-default policy engine for AI agent governance in clinical settings. Enforces access control with SHA-256 audit trail, ensuring all AI agent actions in healthcare contexts are logged, auditable, and compliant with privacy regulations.
Health Asset Value & Exchange Network — a protocol specification for patient data sovereignty. Defines consent management, data provenance tracking, and value exchange mechanisms to give patients control over their health data and enable ethical data sharing.
Patient Scenario Definition Language — a deterministic, portable DSL for specifying patient scenarios in clinical research and decision support. Enables reproducible simulation of patient trajectories across different EHR systems and research environments.
Companion tooling for PSDL. Validates specification files, parses semantic rules, and constructs dependency DAGs to help researchers debug and verify their patient scenario definitions before deployment.
AI-powered digital twin system for surgeon assistance. Provides real-time decision support in the operating room by modeling patient physiological state and predicting intraoperative risk.
Technical Solutions Lead. Built real-time data streaming pipeline using EpicOnFHIR, Python, MiniKube, and Strimzi (Kafka Bridge) to monitor patients at risk of acute kidney injury. System processes 20,000+ FHIR messages/day from the UF Health EHR, enabling timely clinical alerts.
Technical Solutions Lead. Engineered ETL pipelines for a 14M+ encounter multimodal dataset (EHR, clinical images, notes, waveforms). Built OHDSI data quality dashboard, SDoH & Environmental database with PostgreSQL + Geocoding, and EHR data linkage toolkits using React and Docker.
Key Technical Lead. Deployed pervasive sensing system in UF Health ICU using IoT sensors (ambient noise, light, motion) with a real-time web dashboard. Optimized ETL pipeline achieving 45% speed increase for continuous patient monitoring data.
Key Technical Lead. Deployed NVIDIA Flare federated learning platform across multiple hospital sites for privacy-preserving surgical risk prediction. Built XAI-MySurgeryRisk web application in React to surface explainable AI insights to surgeons pre-operatively.
Technical Lead. Built a healthcare digital twin using NVIDIA Omniverse. Integrated environment sensing via IoT and NVIDIA Holoscan SDK. Developed AI dialogue system combining GPT-4, LLaMA 3, and Vicuna for immersive clinical training simulations.
WeChat mini-program for UF campus ridesharing. Built with Dart/Flutter and backend APIs to facilitate carpooling among students.
NLP research at UF Li Lab (NSF Center for Big Learning). Improved voice recognition accuracy from 65% to 78% using deep learning approaches for medical terminology.
Patented hardware/software module for fingerprint recognition. CHN Patent ZL 2016 20390554.1 (May 3, 2016). Developed during undergraduate research.
Full-stack surgical risk prediction platform. PHP/JS/CSS frontend, Java RESTful APIs, Spark/Kafka AI engine. Achieved 1000% caching performance improvement with Memcache/Redis. Later rebuilt with React + Flutter.