Machine Learning Computer Science

TECHNICAL SKILL

EDUCATION

PERSONAL PROJECT

SUMMARY

Languages: Python, C/C++,Numba CUDA, PyTorch, NumPy AI/ML: Deep Learning, CNN Architectures,Transformers Architectures, Model Optimization GPU Computing: CUDA kernel design, GPU memory management, parallel computing Algorithms: Data Structures, Algorithm Analysis, Performance Optimization Language: English (TOEIC 825)

Bachelor of Computer Science Sep 2021 - Sep 2025

University of Science, Vietnam National University Ho Chi Minh City GPA: 3.5/4.0

SqueezeNet from Scratch with CUDA Acceleration Jun 2025 - Sep 2025 Implemented forward & backward propagation of SqueezeNet architecture entirely in CUDA, achieving 92% accuracy on Tomato Disease dataset.

Optimized inference speed to 0.02s/image – 1000x faster than equivalent NumPy implementation. Managed GPU memory allocation, designed parallel kernels for convolutional layers, and reduced latency via shared memory & coalesced access.

Demonstrated deep understanding of CNN mechanics, GPU architecture, and performance bottlenecks in deep learning.

Recent Computer Science graduate from University of Science, VNU-HCM with a strong foundation in AI, GPU computing, and algorithm design. Proven ability to build deep learning models from scratch using CUDA for high- performance inference (1000x acceleration vs NumPy). Passionate about optimizing neural networks and leveraging GPU architecture for real-time AI applications.

NGUYỄN THÀNH LUÂN

AI, MACHINE LEARNING INTERN/FRESH

Address:

Phone:

Email:

Github:

Tan Binh, Ho Chi Minh City, Vietnam

036*-***-***

**************@*****.***

https://github.com/imthanhluan203

CERTIFICATION

TOEIC: 825 (2025)

Sino-Nôm Image Classification using Vision Transformer Jan 2025 - Apr 2025 Developed an automated classification system for Sino-Nôm/Chinese historical document images, categorizing them into general (e.g., books, personal texts), administration (e.g., decrees, official papers), and scene (e.g., inscriptions on monuments) using Vision Transformer (ViT). Collected and curated a dataset of 29,377 images from sources like the National Hán-Nôm Library and online repositories.

Fine-tuned a ViT-Base (patch16-224) model with PyTorch and Hugging Face Transformers, achieving up to 99.7% accuracy and F1-score on test sets.

An Integrated System for E-Commerce Review Analysis and Insight Generation Sep 2024 - Dec 2024 Developed the "ABSA-NLP-Pipeline" project, an integrated Streamlit application to automate the extraction, analysis, and visualization of insights from product reviews sourced from popular platforms Amazon, eBay, and YouTube, including automated collection of product reviews by providing product links. Implemented automated data collection by scraping product reviews from provided product links, then processed and transformed raw review data into actionable intelligence using advanced NLP techniques including AI-powered summarization with OpenAI and a hybrid Aspect-Based Sentiment Analysis (ABSA) approach using RoBERTa and OpenAI.

Contact this candidate