Maring in conputer science, especially, computer vision, NLP

Nguyen Xuan Linh

+ HCM # **********@*****.*** 096*-***-*** ð nguyenxuanlinh § xlinh2301

Summary

A passionate and enthusiastic Computer Science student at University of Information Technology (UIT) - VNUHCM, spe- cializing in Computer Vision. Seeking an internship opportunity to apply and expand technical knowledge. Proficient in Python, with skills in Machine Learning, Deep Learning, Natural Language Processing, and Computer Vision technologies. Eager to contribute to real-world projects and gain hands-on experience in a dynamic environment. Education

BS University of Information Technology (UIT) - VNUHCM, Computer Science

• GPA: 3.2/4.0

Sept 2022 – Sept 2025

Projects

Multimodal Video Retrieval

• Leadership: Leadingafive-member teamtooverseeprojectdevelopmentandcol- laboration.

• System Development:

– CLIP FAISS: Built a systemusingCLIPincombinationwithFAISStoextractfea- tures from images and text, and implemented an efficient search mechanism based on these features. Created an indexing system and designed search al- gorithms to improve search speed and accuracy.

– Scene Text System: Developed a robust Scene Text system by integrating DB- Net for text detection and VietOCR for text recognition. Fine-tuned and opti- mized the models to enhance accuracy and performance.

– Object Detection ASR: Implemented object detection using YOLOv8 to ac- curately detect objects in images. Integrated Whisper for automatic speech recognition (ASR) to convert speech into text with high accuracy.

– Interface Development: Developed the backend interface using FastAPI, en- suring a smooth and responsive user experience.

github.com/repo 2

Hybrid Vehicle Detection Using YOLO and Co-DETR Ensembles

• Team of 3 members, developed a robust hybrid vehicle detection system combin- ing YOLO for fast detection and Co-DETR for improved precision in complex sce- narios.

• Used YOLO11, YOLOv8, and Co-DETR models for detecting and classifying hybrid vehicles in various conditions.

• Utilized techniques like pseudo-labeling for data refinement, and data augmenta- tion to simulate real-world lighting and weather conditions.

• Implementedpost-processingstepssuchaspolygonfilteringandconfidencescore balancing to enhance detection results.

• Achieved **Top 6** in the private test of BKAI Naver 2024 Vehicle Detection com- petition.

github.com/repo 2

Multimodal Sarcasm Detection on Vietnamese Social Media Texts

• Collaborated in a team of 5 to design a multimodal sarcasm detection system for Vietnamese social media texts, leveraging both image and text data.

• Preprocessed text and image data, then extracted features using CLIP, ViT5, and ViLT. Appliedmid-fusionbyconcatenatingtheextractedfeaturesintoaunifiedrep- github.com/repo 2

Page 1 of 2

Last updated in January 2025

resentation.

• Addressed class imbalance using SMOTE (Synthetic Minority Oversampling Tech- nique) and implemented SVM (Support Vector Machine) for prediction, achieving the highest performance.

• Experimented with ensemble techniques by combining different algorithms and explored deep learning models, though SVM consistently delivered the best re- sults.

• Secured **Top 3 public test placement** in the competition. Enzyme Substrate Classification

• Dataset: Kaggle - Playground Series S3E18 - Enzyme Substrate Classification

• Features:

– EDA: Univariate, Bivariate, and Multivariate analysis

– Prediction: MakepredictionsforEC1andEC2usingpre-trainedXGBoostmod- els

– SHAP analysis for model interpretability and feature importance

– Data Preprocessing: Outlier removal, normalization, SMOTE for balancing github.com/repo 2

Technologies

Languages: Python, SQL, C++

Databases: SQL Server, MySQL, ElasticSearch, Kibana Libraries Frameworks: NumPy, Pandas, Scikit-Learn, Keras, TensorFlow, PyTorch, OpenCV, React.js, Node.js, FastAPI Version Control Containerization: Git, GitHub, Docker Page 2 of 2

Contact this candidate