Thái Hoài An

Participated in the AIO25 program to strengthen fundamentals and applied skills in AI/ML through structured modules and practical exercises.

Computer vision project on BUSI ultrasound images comparing multi-task vs sequential pipelines. Implemented U-Net with EfficientNet encoder and evaluated segmentation (Dice/IoU) alongside classification performance.

Faculty-level research project on Vietnamese fake news detection using ReINTEL. Benchmarked BiLSTM, PhoBERT (frozen/fine-tuned), and LLM prompting; reported Accuracy/Macro-F1/AUC and performed error analysis and deployment trade-off review.

End-to-end data analytics project on Vietnam weather data: cleaning, visualization dashboards, and ML-based forecasting/classification. Delivered a Streamlit demo to explore patterns and model outputs.

Data Mining project building a multi-source analytics pipeline for Vietnamese banking stocks. Performed feature engineering for time-series, trained models for return regression and risk forecasting, and summarized insights for investment decision support.

NLP project delivering an end-to-end IE workflow for Vietnamese medical text, including dataset labeling (Label Studio), model training, and evaluation. Explored semi-supervised hybrid relation extraction to improve performance under limited annotations.

Collaborated with a cross-university team to build a logistics solution combining data management, ML allocation, and IoT warehouse tracking to reduce food waste.

MeetMate is an AI-powered meeting intelligence platform developed during the VNPT AI Hackathon. The project addresses a common enterprise pain point: meetings consume significant time but often fail to produce clear decisions, accountable action items, and traceable outcomes. MeetMate integrates automatic speech recognition, real-time context understanding, and post-meeting summarization to transform unstructured conversations into structured minutes, decisions, and follow-up tasks. The solution focuses on enterprise readiness, emphasizing accuracy, explainability, and workflow integration rather than simple note-taking.

Course research project exploring evolutionary optimization for network flow. Built a visual simulator and enhanced genetic operators for faster, more stable convergence on directed graphs.