Building transparent, data-driven intelligence for social impact.
🏆 Published at AJCAI 2025: My paper on Proactive Air Quality Forecasting has been accepted by Springer!
🎙️ New Research: Achieved 33.09% WER on Vietnamese ASR using a custom 5.4M param CNN-BiLSTM model.
✍️ Blog: Wrote about my journey into AI — Read here.
I am a second-year Artificial Intelligence major at FPT University (GPA: 8.0/10.0) and the Founder of SpeedyLabX, a student-led research lab bridging academic theory with real-world application.
📝 Note: I publish research under the international name Gia-Hung Nguyen Le.
My work lies at the intersection of Deep Learning, Time-Series Forecasting, and Explainable AI (XAI). I don't just train models; I build end-to-end systems that are interpretable and deployable.
- 🔭 I’m currently working on: MERR-GAT: fusing Wav2Vec 2.0 (Audio) & RoBERTa (Text) via Graph Attention Networks for explainable emotion recognition.
- 🌱 I’m currently learning: RAG Agents (NVIDIA), Advanced MLOps.
- 💬 Ask me about: Transformer Architectures, SmokeNet, or how to lead a student research team.
| Project | Role | Architecture/Tech | Impact/Status |
|---|---|---|---|
| SmokeNet: Air Quality Forecasting | Lead Author | Transformer, SHAP, Integrated Gradients | ✅ Published @ AJCAI 2025 (Springer) Reduced MAE by 35.9% vs XGBoost. |
| End-to-End Vietnamese ASR | Co-Researcher | CNN-BiLSTM (Residual), CTC Loss | 🔄 In Progress Trained on 745 hours of speech. Achieved 15.28% CER with only 5.4M params. |
| MERR-GAT: Multimodal Emotion Recognition | Co-Lead / Corresponding Author | Wav2Vec 2.0, RoBERTa, GATv2 | 🧪 In Progress Novel fusion of acoustic & textual features via Dynamic Graph Attention Networks for explainability. |
Languages 
AI & Data Science 
Tools & Platforms 
- 📜 NVIDIA: Building RAG Agents with LLMs
- 📜 IBM: Full Stack Software Developer Professional Certificate
- 📜 IBM: Data Science Fundamentals with Python and SQL
