Kaiyuan Wu

About

I am a 2nd-year PhD student in Computational Biology and Bioinformatics (CBB) at Duke University School of Medicine and a concurrent MA student in Applied Ethics and Policy with a Tech Ethics & Policy concentration. My work is positioned at the intersection of clinical AI & (multi-)agentic framework, AI safety, monitoring, and governance.

The central question guiding my work is how clinical AI systems can become reliable enough to support care delivery, auditable enough to be governed, and interoperable enough to be translated across real healthcare environments.

Education

Research Mission

My proposed research endeavor is to develop, validate, and translate safety and governance frameworks for high-risk clinical artificial intelligence systems in U.S. healthcare. The work focuses on agentic clinical decision support, FHIR-based care planning, multimodal clinical prediction, continuous monitoring, and risk-control mechanisms for real-world deployment.

The practical need is clear: as clinical AI increasingly moves into discharge planning, critical care support, imaging, disease trajectory prediction, and hospital workflow optimization, model performance alone is not enough. AI systems must be clinically aligned, monitored after deployment, resilient to failure modes, and governed in ways that protect patients while supporting clinicians.

The broader value of this mission is system-level. Safe clinical AI governance is not limited to one model, one hospital, or one employer. Methods for validation, monitoring, bias detection, risk control, and workflow integration can support and scale across hospitals, research institutions, public health settings, and healthcare technology platforms that need trustworthy AI infrastructure.

In this sense, my work supports the safe translation of clinical AI from research prototypes into governable, monitorable, and clinically reliable healthcare infrastructure.

Research Focus

These research areas are organized around a single objective: building the technical and governance layer between AI innovation and responsible clinical deployment.

Clinical AI safety and governance

Evaluation, monitoring, and governance frameworks for high-risk AI systems in clinical settings, including asymmetric risk evaluation, runtime safety surveillance, audit design, and post-deployment oversight.

Agentic clinical decision support

Multi-agent systems for discharge planning and critical care decision support, with emphasis on guideline recall, self-improvement, counterfactual simulation, clinician-in-the-loop design, and failure-aware planning.

Multimodal clinical prediction

Models that integrate EHR, imaging, text, waveforms, and multi-site clinical datasets for trajectory prediction, risk identification, and care-planning support.

Publications and Preprints

1. Planner-Auditor Twin: Agentic Discharge Planning with FHIR-Based LLM Planning, Guideline Recall, Optional Caching and Self-Improvement.
   Wu K, Nagori A, Kamaleswaran R. arXiv preprint arXiv:2601.21113, 2026. Preprint
2. CXR-TFT: Multi-modal Temporal Fusion Transformer for Predicting Chest X-Ray Trajectories.
   Arora M, Ali A, Wu K, Davis C, Shimazui T, Alwakeel M, Moas V, Yang P, et al. International Conference on Medical Image Computing and Computer-Assisted Intervention (MICCAI), 2025.
3. The detection, function, and therapeutic potential of RNA 2’-O-methylation.
   Wu K, Li Y, Yi Y, Yu Y, Wang Y, Zhang L, Cao Q, Chen K. The Innovation Life, 3(1), 100112, 2024.
4. The therapeutic potential of targeting the CHD protein family in cancer.
   Zhang M, Wu K, Zhang W, Lin X, Cao Q, Zhang L, Chen K. Pharmacology & Therapeutics, 256, 108610, 2024.
5. Exploring hippocampus segmentation on unbalanced data set using U-Net-based models.
   Wu K. Applied and Computational Engineering, 6, 758–763, 2023.
6. The effect of selective c-MET inhibitor on hepatocellular carcinoma in the MET-active, β-catenin-mutated mouse model.
   Zhan N, Michael AA, Wu K, Zeng G, Bell A, Tao J, Monga SP. Gene Expression, 18(2), 135, 2018.

Translation, Intellectual Property, and Governance

Intellectual property

• U.S. Provisional Patent Filed: OCA-Sentinel Framework for Trustworthy Multi-Agentic AI. U.S. Provisional Application No. 64/030,784, Systems and Methods for Multi-Agentic AI. Duke University OTC Ref. T-008960. Co-inventors: R. Kamaleswaran, A. Nagori, K. Wu. Status: Patent Pending. Sponsored by NIH GM139967.
• Co-inventor: provisional patent for a VA-ECMO multi-agentic clinical decision-support framework.

Grant contributions

• NIH R01 Proposal (2026): co-investigator on a VA-ECMO multi-agent decision support R01; responsible for hierarchical multi-agent architecture and the safety evaluation plan.

Governance and stakeholder engagement

• Contributing PhD student in the drafting of Duke University’s Agentic AI Monitoring guideline document, a governance and monitoring initiative led by Dr. Eric Poon at Duke Health.
• Engaged in translational discussions with Duke Health system leadership, Duke Heart Center leadership in an ARPA-H-adjacent context, and Duke Clinical and Translational Science Institute (CTSI) rural health implementation discussions focused on equitable AI deployment in underserved communities.

Technical Skills

• Programming and ML: Python, R, PyTorch, deep learning frameworks, agentic AI architectures, LLM-based systems, X-AI, trustworthy & responsible AI.
• Policy and governance: AI safety & governance, FDA & NPMA policy & regulatory landscape (US & China), clinical bioethics, technology ethics.
• Web and software: Full-stack web application development, open-source software engineering, DevOps.
• Clinical data: EHR data systems, MIMIC-IV, eICU, multimodality, SNOMED-CT, medical ontology, multi-site clinical data integration.
• Bioinformatics: Nanopore sequencing, FastQC, HISAT2, SAM/BAM, ChIP-seq analysis, RNA modification detection, gene expression analysis.
• Wet lab: Magnetic nanoparticle synthesis, stem cell culture, transfection, H&E staining, protein extraction, plasmid manipulation.

Contact

Email: vincent.wu [AT] duke.edu
Institution: Duke University School of Medicine, Durham, NC