I am a Ph.D.-trained computational scientist working at the intersection of physics-based modeling, machine learning, and industrial engineering.
My background spans computational fluid dynamics, high-performance computing, structural aerodynamics, and reproducible scientific workflows. More recently, I have led industrial R&D work involving AI-based visual inspection, data pipeline design, and cross-functional product deployment.
I am now building on that foundation through hands-on work in PyTorch, computer vision, edge deployment, FastAPI, Docker, AWS, and LLM-enabled applications.
- 🎓 Ph.D.-trained computational scientist with an engineering background
- 🔬 Experience in CFD, HPC, numerical simulation, and experimental validation
- 🏭 Industrial R&D experience in AI visual inspection and monitoring systems
- 🤝 Experience translating client workflows into technical requirements and repeatable data pipelines
- 🤖 Practical work in PyTorch, CNNs, PINNs, model deployment, and LLM applications
- 🌏 Bilingual technical communication in English and Chinese
- Led cross-functional R&D projects for AI-based visual inspection and industrial monitoring systems.
- Worked with industrial clients to define product requirements, data collection processes, and image-annotation standards.
- Developed CFD workflows using OpenFOAM, ANSYS Fluent, Python, Linux, and HPC environments.
- Explored hybrid engineering workflows combining numerical simulation, physics-informed neural networks, and AI surrogate models.
- Conducted doctoral research on tornado-like vortices, wind loads, and structural aerodynamics.
A deployment-oriented computer vision workflow covering dataset preparation, EfficientNet fine-tuning, evaluation, model compression, ONNX export, INT8 quantization, and latency-focused edge deployment, backend flow with FastAPI and deployment with Docker.
Python PyTorch EfficientNet ONNX INT8 Quantization FastAPI Docker
An in-progress application for extracting structured line items from handwritten receipts, recalculating totals, and presenting discrepancies for human review. The project explores multimodal LLM and OCR-based approaches while prioritizing numerical reliability.
Python Multimodal LLMs OCR Structured Data Validation Logic
A project-based learning record focused on convolutional neural networks, image-classification workflows, and practical PyTorch development.
Python PyTorch CNNs Computer Vision
- Computer vision for industrial inspection
- Efficient and deployable PyTorch models
- Physics-informed neural networks and surrogate modeling
- Edge inference and model optimization
- AI agents and LLM-enabled applications
- Reproducible ML services with FastAPI and Docker
| Area | Technologies & Methods |
|---|---|
| Programming & Tools | Python, Git, GitHub, Linux, MATLAB |
| Machine Learning | PyTorch, CNNs, transformer-based models, PINNs, surrogate models |
| AI Engineering | Model training and evaluation, ONNX, edge deployment, FastAPI, Docker, AWS |
| Scientific Computing | CFD, OpenFOAM, ANSYS Fluent, HPC, numerical simulation |
| Data & Visualization | Scientific data analysis, time-series analysis, post-processing, 3D visualization |
| Collaboration | Technical documentation, mentoring, client communication, cross-functional coordination |
I use structured coursework to extend my computational science and industrial R&D background into an applied AI engineering skill set. My recent learning covers machine learning, deep learning, PyTorch, computer vision, AI agents, backend development, and containerized deployment.
View my complete certification portfolio →
I am interested in roles and collaborations involving:
- Applied machine learning and computer vision
- Scientific machine learning and AI for engineering
- Industrial AI and visual inspection
- Computational science and research software
- Data Science and Analysis
- Model optimization, deployment, and edge AI
- GitHub: @Isaiah1001
I believe the strongest AI systems combine sound engineering judgment, reproducible implementation, and a clear understanding of the real-world problem they are designed to solve.