Kam Kiu Aluminium Group

• Built and deployed an automated OQC reporting system with PyQt, FastAPI, and Celery/Redis, reducing manual processing by 95% and saving 6+ hours/day, across multiple projects and teams
• Built a real-time manufacturing performance dashboard with PyQt, enabling live production monitoring and faster operational decisions
• Designed and deployed a shared Nginx reverse-proxy infrastructure for multiple Dockerized applications on Linux servers, centralizing HTTP/HTTPS routing and backend access while maintaining application isolation
• Engineered and deployed a full-stack training management platform with Next.js, FastAPI, and self-hosted MinIO object storage, providing centralized management of training content and media resources
• Engineered a full-stack assessment platform that digitised a paper-based examination workflow using Next.js, NestJS, and LLMs, automating question-bank parsing, assessment generation, and free-text answer grading, with DingTalk integration for employee information and Langfuse for LLM observability and evaluation

Resume LLM Assistant

• Built a resume chat assistant for HR professionals to query candidate resumes using natural language, leveraging a Next.js frontend and FastAPI backend
• Designed and implemented a resume ingestion pipeline using Alibaba text embedding models to generate semantic embeddings for resume sections
• Integrated ChromaDB vector storage for efficient similarity search and semantic retrieval of candidate information
• Developed a custom Retrieval-Augmented Generation (RAG) pipeline with LangChain to deliver context-aware responses from relevant resume sections
• Implemented semantic search workflows enabling recruiters to quickly retrieve candidate skills, experience, and qualifications from unstructured resume data

Personal Website

• Engineered an interactive 3D homepage with React Three Fiber and Blender, featuring interactive objects, camera views, and navigation
• Developed a full-stack activity telemetry system with a desktop client-side logger that records keyboard/mouse usage in MongoDB, with interactive usage dashboards on Next.js frontend
• Created an interactive knowledge graph using an Obsidian plugin to extract note metadata into MongoDB, visualized with PixiJS/WebGL and D3 force-directed simulation
• Implemented a Claude Code hook to capture per-request LLM usage, including token counts, cost, and model, with a Next.js dashboard for usage analysis
• Deployed the personal website on Vercel with automated production deployments