AI undergraduate researcher at FPT University, Ho Chi Minh City, Vietnam.
GPA 8.62/10 · Expected graduation: December 2027.
I work on time-series forecasting and computer vision, with a focus on explaining model predictions. I founded SpeedyLabX, where we read papers, build experiments and work through research questions together.
Open to research internships and AI/ML engineering opportunities.
Portfolio · Research CV · Google Scholar · Email
Proactive Air Quality Forecasting and Health Alert System for Melbourne — Springer LNAI. First & corresponding author.
Gia-Hung Nguyen Le, Gia-Bao Pham Hoang, Tan-Phat Vo, Thu Le, Nhu Nguyen.
I designed the transformer model and chronological evaluation pipeline, then used attribution and counterfactual analysis to inspect its forecasts. The model uses 48 hours of history to forecast the next day's PM₂.₅.
Paper · Research code · Methods & evaluation
Four manuscripts under review
- ACIIDS 2027: Auditing Multi-Source Cyanobacterial Bloom Forecasts under Sparse Environmental Telemetry. First & corresponding author.
- AISTATS 2027: Distribution-Free Coverage of Attack Classes on Devices Absent from Training. First author. Code.
- AJCAI 2026: Warning Before the Evening Peak: Health-Centred Air Quality Alerts in New Zealand and Australia. Second author.
- ACM SIGKDD 2027: Trustworthy Multimodal Anomaly Detection in Time Series with Conformal Calibration and Counterfactual Attribution. Second author.
| Project | Work & evidence |
|---|---|
| MERR-GAT | Audio/text emotion recognition with WavLM + Compact Conformer, BERT/RoBERTa and GAT. Second & corresponding author; in progress. |
| Vietnamese ASR | Co-researcher on a CNN + residual BiLSTM / CTC system. Reported 33.09% WER, 5.4M parameters, on a 745-hour Vietnamese corpus. |
| PPE Detection | Built a nine-class YOLOv8 pipeline with webcam input, compliance logic and a Streamlit interface. Course project. |
- FCAJ Cloud Intern / NutriTrack: Built recipe-generation and coaching workflows using Bedrock, Lambda, DynamoDB and AppSync. Implementation & role.
- BioListen VN: UI and LLM integration with Team NeuraX.ai. 6th in Round 1 of Vietnam AI Innovation Challenge 2026. Case study.
- CosmicOptic: Frontend and backend for a NASA Space Apps 2025 prototype. The web demo uses synthetic signals. Case study.
Python, SQL, PyTorch, TensorFlow, Hugging Face, scikit-learn and OpenCV. For evaluation and explanations: SHAP, Integrated Gradients, DiCE and conformal prediction. For applications: FastAPI, Streamlit, React, TypeScript and AWS tools.
Project-linked toolkit · Writing
Away from the screen, I like trail running, the gym and photography. I also have a soft spot for ducks.
LinkedIn · ORCID · Scopus · Web of Science