At THE GIOI DI DONG INFORMATION TECHNOLOGY LIMITED COMPANY, I developed a real-time product recognition system covering approximately 3,000 products. I combined YOLO detection, image preprocessing, crop-and-reclassify pipelines, and embedding-based retrieval to handle different product appearances and camera viewpoints.
I optimized GPU inference with ONNX, TensorRT, FP16/INT8 optimization, and batch processing, reducing latency from approximately 1,500 ms for a single image to 300 ms for a batch of three. I deployed inference services across NVIDIA GPU and embedded environments.
I also developed enterprise document and knowledge retrieval workflows, using MongoDB, vector search, and structured queries for heterogeneous business information. For an AI candidate matching project, I built pipelines that represented resumes and job descriptions semantically and ranked candidates by job-relevant similarity.
