AI/ML researcher building systems for biomedical discovery — from fine-tuning vision-language models for radiology to physics-informed neural networks for cancer simulation.
I'm an undergraduate researcher and a B.S. Mathematics (Conc. Computer Science) student at Georgia State University. My work sits at the intersection of AI/ML and healthcare, where I build computational models and data-driven systems to solve complex biological problems. I'm currently exploring opportunities for Summer 2026 internships in AI/ML, data science, and computational biology.
🔭 I'm currently fine-tuning MedGemma vision-language models for automated mammography report generation and clinical triage.
🧠 I'm deepening my expertise in PyTorch, LoRA/QLoRA, and multimodal deep learning architectures.
🔬 My research spans ML pipelines for cellular aging analysis, PINNs for cancer invasion, and neuromorphic biocomputing.
📫 How to reach me: vanthienphan2004.work@gmail.com
- Engineered a system to automatically classify and prioritize high-risk mammography cases using a fine-tuned Google MedGemma 1.5 4B-it vision-language model with LoRA and 4-bit quantization.
- Achieved high-fidelity report generation: ROUGE-L: 0.6693, METEOR: 0.7187, Word-Level F1: 0.6789.
- Developed an automated, config-driven 7-step end-to-end ML pipeline for RPE cell classification.
- Extracted 133 morphological/texture features and achieved 90%+ cross-validation F1-score using a stacking ensemble (XGBoost, LightGBM, CatBoost) with a Logistic Regression meta-learner.
- Engineered a PINNs model to simulate collective cancer invasion as a data-efficient, mesh-free solution for complex PDEs.
- Utilized Tensorized Fourier Neural Operators (TFNO) and explored Seq2Seq PINO and Augmented Lagrangian methods to improve solution accuracy.
- Simulating biological "wetware" logic gates using FitzHugh-Nagumo and Hodgkin-Huxley models with phase space trajectory analysis and coincidence detection.
