<p>Nowadays, due to the highly competitive job market, it has become extremely difficult for employers to efficiently select the most suitable candidates from a large pool of job applications. Recruiters invest significant amounts of time and effort in manually categorizing and organizing resumes against available vacancies. With the advancements in machine learning, natural language processing, and the emergence of novel word embedding techniques, new opportunities to streamline and automate recruitment workflows. In this paper, we propose an innovative automated resume classification system designed to exploit these advancements. The system begins by converting heterogeneous data into a consistent textual format. Next, we applied text data augmentation techniques to enhance the generalization ability of our system and effectively address class imbalance. Subsequently, advanced word embeddings are employed, with Llama3.1 demonstrating superior semantic and contextual performance compared to TF-IDF and DistilBERT. These embeddings are then fed into classifiers to perform the classification. Moreover, we constructed a dataset of 1,500 resumes collected from multiple sources, including LinkedIn, Kaggle, recruitment agencies, and professional contributions, spanning five distinct information technology domains. Experimental results demonstrate that models using the Llama3.1 embedding technique achieved higher accuracy than those employing TF-IDF or DistilBERT embeddings. In particular, the Logistic Regression/Llama3.1 model with text augmentation obtained the highest accuracy of 0.99%, significantly outperforming the other models</p>