Tai Wong

CS + BioE @ UIUC

Tai Wong

Building at the intersection of AI and biology.

Resume

About

I'm a Computer Science and Bioengineering student at the University of Illinois Urbana-Champaign. My research spans deep learning for biomedical imaging at OHSU and computational genomics at UCSF. Outside of research, I'm interested in reinforcement learning and getting computers to play games.

Research

2025 — Present

Research Assistant · Dream Lab, UIUC

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.

PythonLangGraphNeo4jLLMsMulti-Agent Systems
2023 — 2025

Deep Learning Research Intern · COOL Lab, OHSU

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.

PythonPyTorchCNNU-NetComputer VisionMATLAB
2022 — Present

Research Assistant · Mattis Lab, UCSF

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.

PythonPubMedBERTNLPBioinformaticsPostgreSQL
View Full Resume

Projects

Featured projects you should check out.

MeshNL

Multi-stage hierarchical MeSH classification trained on the full PubMed baseline. Stage 1 fine-tunes BiomedBERT for 15-branch prediction (macro recall >0.95); Stage 2 fine-tunes a BioLORD dual encoder on 3M paper-term pairs with hard negative mining for fine-grained term retrieval (Recall@50). Engineered a streaming MeSH XML parser and Colab training infrastructure with mid-epoch checkpointing.

PythonPyTorchBiomedBERTBioLORDHuggingFace Transformers

PyEntrez

Python wrapper for NCBI's EDirect command-line tools. Provides a programmatic API for searching PubMed and other NCBI databases, fetching records, discovering cross-database links, pulling citation counts, and querying by MeSH terms. Includes batch processing with parallel workers for bulk operations.

PythonNCBI EDirectPubMedMeSH

Here's some cool projects I worked on in no particular order:

SkyHub

Full-stack Hypixel SkyBlock stats platform (successor to SkyStats, 15,000+ MAU): 42-endpoint FastAPI backend with PostgreSQL and background workers for leaderboard ranking and networth calculation across 40+ item upgrade types. Self-hosted on a Mac Mini via Docker Compose with Cloudflare Zero Trust tunnel, launchd auto-start, and automated 6-hour PostgreSQL backups.

FastAPIPostgreSQLDockerCloudflare TunnelSwiftUI

IllinoisBuddy

UIUC course planning web app aggregating Rate My Professor ratings, GPA distributions (Wade Fagen dataset), and Course Explorer API data. Flask + Google OAuth backend with React/TypeScript drag-and-drop multi-semester planner, GPA visualization, and schedule conflict detection. Migrating to PostgreSQL + FastAPI for production.

FlaskReactTypeScriptTailwindSQLitePostgreSQL

SkyStats

Hypixel SkyBlock stats tracker with 15,000+ MAU, built and maintained over three years. Rewrote the iOS client to SwiftUI and migrated the backend to MongoDB, reducing latency by 20%.

SwiftUIMongoDBPython