Đỗ Hoàng Minh dom587316

🐣 Đỗ Hoàng Minh (dom587316)

👋 Hi, I'm Đỗ Hoàng Minh (he/him)
🎓 Student at Vietnam National University, Hanoi (VNU - UET)
🏠 Based in Hanoi, Vietnam
📍 Hometown: Lý Nhân, Hà Nam, Vietnam
💼 Aspiring AI Engineer & Deep Learning Researcher
❤️ Interested in Computer Vision, MLOps, LLMs, and High-Performance Inference
💥 PyTorch | FastAPI | Docker | Qdrant

An end-to-end, high-performance license plate recognition system utilizing a layered microservices architecture.

High-Performance AI: Custom-trained YOLOv5n/s models for license plate detection and character recognition (OCR) achieving 94.69% accuracy on a 471-sample dataset.
Microservice Architecture: Packaged with 9 Docker containers including FastAPI, React Frontend, Redis Queue, RQ Worker, PostgreSQL, MinIO, Qdrant Vector DB, Prometheus, and Grafana.
Fuzzy Search Integration: Leverages Qdrant Vector DB to provide similar plate suggestions when OCR has minor character recognition errors.
Low Latency & Scalability: Designed an async background queue using Redis/RQ. Transitioned to a Warm Cache SimpleWorker strategy reducing inference latency to 100-200ms.
Real-Time Stream Optimization: Implemented a FrameTracker (IoU & Centroid tracking) at the backend to bypass redundant OCR calls for stationary vehicles, saving 90% CPU/GPU resources.
Full Observability: Custom Prometheus metrics monitored through Grafana dashboards tracking queue depth, inference latency, and API requests.

An interactive simulation of a Micromouse robot solving a maze using classic algorithms.

Algorithm Design: Implementations of classic Floodfill Algorithm for maze solving.
Hardware Integration: Arduino implementation to run on real Micromouse physical robots.