Sydney, New South Wales, Australia
Machine Learning Engineer with 5+ years building production-grade generative AI systems for medical imaging and enterprise applications. PhD graduate from the University of Sydney, specializing in deep learning architectures that process large-scale 3D imaging, text, and numerical data. Recent Impact: •RAG Systems: Built a domain-specific RAG system for oncology guidelines using FastAPI, FAISS, and LangChain, with automated ingestion pipelines and real-time retrieval in an interactive Next.js prototype •GenAI models for Medical Image Generation: Developed a full-resolution 3D generative inference pipeline for ultra-low-dose PET reconstruction, achieving +2.0 dB PSNR improvement and 20% lower error rates across multi-scanner datasets, while reducing GPU memory by 18% and inference latency by 20% on NVIDIA A100 •Industrial AI: Developed computer vision systems for multi-view object detection across 7 camera angles, co-authoring 4 patents for large-scale industrial inspection automation. Technical Expertise: •Generative AI: 3D GANs, diffusion models, self-supervised learning for medical image synthesis and denoising •LLMs & RAG: OpenAI API, Claude API, Hugging Face Transformers, vector databases (FAISS), embedding pipelines, semantic search •Production ML: PyTorch-based model development, FastAPI services, OAuth 2.0 authentication, automated deployment pipelines •Cloud & DevOps: AWS, Cloud HPC, containerization, CI/CD automation, system monitoring. Recognition: 🏆 First Place, Ultra-low Dose PET Imaging Challenge (MICCAI 2022) 📊 8+ peer-reviewed publications in IEEE and medical imaging conferences 🎓 PhD from University of Sydney under Prof. Jinman Kim and Prof. Dagan Feng.
I am a Machine Learning Researcher specializing in Generative AI, Large Language Models (LLMs), and multi-modal deep learning, with a strong focus on healthcare and medical imaging applications. Key Projects & Contributions: *LLM & RAG Systems: •Developed a domain-specific RAG system for oncology guidelines using FastAPI, FAISS, and LangChain, with automated text ingestion, chunking, and embedding pipelines •Built an interactive Next.js prototype demonstrating real-time semantic search, evidence display, and context-restricted Q&A •Integrated secure API services with OAuth 2.0 and token-based authentication for production readiness *GenAI for Medical Image Generation: •Designed a hybrid CNN–MLP architecture for full-resolution 3D PET reconstruction from ultra-low-dose scans, achieving +2.0 dB PSNR improvement and 20% lower error rates across multi-scanner datasets •Optimized inference for clinical deployment on NVIDIA A100, reducing GPU memory usage by 18% and latency by 20% compared to Transformer baselines •Built self-supervised 3D GANs for PET denoising without paired training data, improving PSNR by +0.5–0.9 dB while maintaining 0.998 SSIM Technical Focus: •Generative AI architectures (GANs, diffusion models, hybrid CNN–MLP networks) •LLM APIs (OpenAI, Claude), vector databases (FAISS), embedding models (SentenceTransformers) •Multi-modal learning combining 3D imaging, clinical text, and structured medical data •Production ML pipelines: PyTorch model development, containerization, CI/CD automation, monitoring
• Developed machine learning models for container instance segmentation and damage detection in industrial inspection systems. • Created an image registration pipeline to enhance multi-view image consistency for reliable inspections. • Optimized production-grade segmentation models to improve detection performance and system robustness. • Contributed to four filed patents related to container inspection systems and algorithmic innovations.
• Developed and implemented a vessel segmentation module, enhancing the vessel recognition pipeline. • Conducted model testing on three advanced video object detection models, optimizing for the best performance. • Managed data collection and cleaning for an ultrasonic image dataset, ensuring high-quality inputs for accurate detection.