In this paper, we study the problem of keyword extraction from IT job descriptions. We develop and present a new annotated dataset, curated specifically for the IT recruitment domain, that includes technical terms and job-related keywords critical for effective talent acquisition. We compare the performance of a fine-tuned distill-roberta model with baseline models KeyBERT and ChatGPT, and demonstrate its superior ability to identify domain-specific terms. Our experiments show significant improvements in recognizing specialized technical skills. We hope our method provides valuable insights into the intersection of AI-driven keyword extraction and recruitment technology.