At VNG Hanoi (Green Node), I co-developed a production document-AI gRPC microservice for computer vision, OCR, and structured information extraction across more than 60 Vietnamese banking and insurance document types. I also improved detection validation IoU from 0.85 to 0.95+ through targeted data augmentation.
I rebuilt the recognition stage around per-line OCR, cutting recognizer calls by about four times, and fine-tuned SVTRv2-GTC for Vietnamese line-level OCR. The work raised Levenshtein similarity from 0.90 to 0.97 across printed and handwritten lines.
At HUST Lab, I’m researching vision-language models for medical image segmentation, including a Text-Guided Frequency Selection module. I’ve also developed projects involving a LangGraph multi-agent system and an AI pipeline for blockchain vulnerability classification.
