Data Engineering Intern Candidate

Duong Tan Hung

Artificial Intelligence Student

Da Nang, Vietnam +84-905-***-*** **************@*****.*** github.com/hungdata Education

FPT University Sep. 2023–Present

Bachelor of Artificial Intelligence Da Nang, Vietnam GPA: 8.7/10

Technical Skills

Programming Python

Databases PostgreSQL, MySQL, MongoDB

Data Analysis & Visualization NumPy, Pandas, Matplotlib, Seaborn Machine Learning Scikit-learn, XGBoost, CatBoost

Deep Learning PyTorch, TensorFlow, Keras

Natural Language Processing Hugging Face Transformers, BERT, LLaMA, TF–IDF Computer Vision OpenCV, YOLO, Grad-CAM

Tools & Platforms Git, Docker, Linux/Ubuntu, Jupyter Notebook, Google Colab, Kaggle Research Experience

Analyzing Resampling Effects on Class Decision Boundaries for Diabetes Prediction Jan.–May 2026 Manuscript submitted to the APWeb-WAIM 2026 Workshop GitHub

• Designed a leakage-aware machine learning pipeline for diabetes prediction using the BRFSS healthcare dataset.

• Evaluated SMOTE, ADASYN, ENN, Tomek Links, and hybrid resampling methods under severe class imbalance.

• Benchmarked Random Forest, XGBoost, Gradient Boosting, and CatBoost, raising minority-class F1 from the low 0.20s to 0.466 with ENN and Gradient Boosting/CatBoost. Unsupervised Domain Adaptation for Lightweight Fruit Freshness Classification Jul. 2026–Present MobileNetV3-Small, PyTorch, UDA GitHub

• Developed a leakage-free domain adaptation benchmark and compared ERM, DANN, CDAN, CDAN+E, DSAN, and DSAN+E.

• Extended DSAN with entropy conditioning, instance reweighting, and Minimum Class Confusion, achieving 90.68% target accuracy with DSAN+E+IR+MCC.

• Increased target accuracy from 80.37% with ERM to 92.79% with CDAN, a 12.42 percentage-point gain, and analyzed domain alignment using t-SNE and Grad-CAM. Projects

Local-to-Global Image Retrieval for World Cup Images Jul. 2026–Present PyTorch, FIRe, CVNet-R101, Graph Diffusion GitHub

• Developed an end-to-end image retrieval system combining FIRe local features, CVNet-R101 global descriptors, Chamfer distance, and graph diffusion.

• Built offline feature extraction and indexing pipelines, with PyTorch mini-batching optimized to run on less than 1 GB of GPU memory.

• Implemented online Top-K retrieval and mAP evaluation, integrated with a FastAPI backend and a responsive Next.js frontend.

Vietnamese Fake News Detection Jan.–May 2026

Python, Scikit-learn, LSTM, BERT, LLaMA 8B GitHub

• Created a Vietnamese fake-news dataset by translating an English dataset with LLaMA 8B and applying post- processing to preserve semantics and class labels.

• Implemented and compared TF–IDF-based machine learning models, LSTM, and BERT for binary news classifi- cation.

• Achieved 98.1% accuracy with BERT, approximately 95% with LSTM, and 90% with a Decision Tree baseline. Languages

English: IELTS Overall 5.5 Japanese: Basic proficiency

Contact this candidate