Tai Wong

Building a multi-agent AI system using LangGraph to identify biomedical research opportunities by analyzing data from PubMed, NIH Reporter, and ClinicalTrials.gov. Architecting a Neo4j knowledge graph with MeSH-based semantic structure for research trend analysis and gap detection. Also researching code-first approaches to improve LLM trustworthiness, where agents design statistical analyses in sandboxed environments to ground claims in verifiable outputs.

Developing an NLP pipeline using PubMedBERT and local LLMs to extract structured cell differentiation protocols from ~300,000 PubMed papers, building a unified database linking source cells, growth factors, and experimental conditions for stem cell research. Also designed bioinformatics pipelines predicting miRNA-mRNA interactions related to NAFLD, contributing to a published paper on regulatory networks in metabolic disease.

Developed CNN and U-Net models to analyze OCTA scans for dilated vessel detection as a biomarker of diabetic retinopathy. Improved segmentation through hyperparameter tuning and data augmentation, evaluating with AUC, IoU, and sensitivity metrics.