CONTACT
• **************@*****.***
• https://github.com/KhanhMinh2004
EDUCATION
UNIVERSITY OF SCIENCE AND
EDUCATION, THE UNIVERSITY
OF DA NANG
GPA: 3.2
Bui Tran
Khanh Minh
AI INTERN
I’m a fourth-year student who is interested in Artificial Intelligent (AI). I really want to join an internship in a dynamic environment to learn, gain experience and improve my skills. I hope I will have an opportunity to show myself TECHNICAL SKILLS
• Languages: Python, Sql
• Developer Tools: Git, Pycharm, VS
Code, Google Colab, Jupyter Notebook,
Ubuntu Linux, Docker
• Image Processing: Augmentation
• Libraries & Frameworks:
- AI/ML: TensorFlow, Keras, Scikit-
learn, Ultralytics YOLO, PaddleOCR,
Pytorch
- Image Processing: OpenCV, Pillow,
Data Augmentation, Matplotlib
- Data Handling: NumPy, Pandas
- GUI & OS: Tkinter, Subprocess, OS,
Time
• APIs & Platforms: Kaggle API, Google
Colab, Google Drive API
LANGUAGES
ENGLISH
• Reading and writing: Intermediate level
• Listening: Intermediate level
• Speaking: Basic level
PROJECT
Real-Time Face Recognition System
Aug. 2024 – Present
• Developed a real time face recognition system using Haar Cascade for detection and custom CNN for classification.
• Achieved 90% accuracy for static faces and 77% for moving faces in real-world webcam conditions.
• Built a full image preprocessing pipeline: face detection, normalization, and live classification.
• The CNN model was trained using real face images. It learned through multiple convolution, pooling and dropout layers and optimized using appropriate algorithms to achieve high performance.
• Github repo: https://github.com/KhanhMinh2004/detect_face_project Real-Time Acne Detection System using YOLOv8
Mar 2025 – Present
• Built a real-time object detection system to identify acne spots using YOLOv8 large model.
• Collected and fine-tuned a pretrained YOLOv8l model on Kaggle acne detection dataset in YOLOv8 format.
• Achieved the detector highly under realistic webcam-based facial conditions.
• Fine-tuned YOLOv8l on over 900 labeled facial images with acne using YOLOv8 format; trained over 100 epochs.
ID Card Information Extraction and Face Verification using PaddleOCR & DeepFace
Jan. 2025 – Present
• Built a Python pipeline to extract Vietnamese citizen ID card data using PaddleOCR.
• Resized ID card images to a fixed resolution (960 720), enduring the card was captured and aligned properly to enable accurate region cropping for OCR extraction (e.g., name, date of birth, ID number).
• Applied DeepFace to verify identity between the ID card photo and a real-world image, using RetinaFace for face detection and ArcFace for face recognition by computing similarity between face embeddings
Fine-tuning LLaMA 3.2 using Unsloth for Efficient Instruction Tuning Nov. 2024 – Present
• Customized and fine-tuned Meta LLaMA 3.2 3B-Instruct using Unsloth for faster and memory-efficient training on consumer GPUs
• Integrated Medical-Sci-Instruct-1M dataset with ShareGPT format for supervised fine-tuning (SFT).
• Implemented instruction-response fine-tuning using trl and Hugging Face Transformers with custom chat templates
• Leveraged 4-bit quantization (bnb-4bit) and parameter-efficient fine- tuning to train 0.81% of parameters on a single GPU with 2x faster speed using Unsloth
ACTIVITIES
• Actively participated in university activities.
• Experienced working in an English-speaking environment.