Ha Trong Nguyen // selected_work
AI products shipped from research prototype to deployed service. Mostly open source, mostly Python, mostly TypeScript when there is a UI.
ANZ Innovation
Full-stack AI educational platform with RAG chatbot, multi-agent grading, and OCR document processing. LLM agent workflows via LangGraph for performance analysis, lesson planning, and quiz generation. GPU-accelerated OCR pipeline with PaddleOCR for document processing.
ANZ Innovation
EdTech AI platform with 5 modules: OCR, Grading, Evidence Analysis, RAG, and Quiz Generation. RAG system with Milvus vector database for semantic search; automated grading with rubric parsing. Deployed as production Docker service with Next.js frontend integration.
Backend API integrating SwingNet AI and MediaPipe Pose for golf swing recognition (8 phases, 93% accuracy). Automated scoring (0–10) with coaching feedback and corrective drill recommendations. Deployed on Hugging Face Spaces.
Led a 3-member team to build OCR-based product information extraction for packaging inspection with 97% accuracy. First Prize in university competition, published at SOICT 2025 (Springer CCIS).
ResNet-Transformer architecture for Vietnamese text recognition, achieving 95% accuracy on VinAI dataset.