At TMA Solutions, I developed and evaluated deep learning models for continuous sign language recognition, building end-to-end pipelines from data preprocessing through real-time inference. The work achieved 20% WER on PHOENIX-2014T and 96% accuracy on a self-collected dataset.
At AiTa Lab - FPT University, I co-developed a lightweight YOLO11 detector for UAV small-object detection and developed a GNN-BiLSTM model for EEG emotion recognition. I’m also developing an AI-powered travel planning system with a multi-agent architecture and a RAG pipeline for personalized itinerary generation.
