The growing use of automated recruitment systems has raised concerns about gender bias in job descriptions. Subtle linguistic cues can discourage qualified candidates from underrepresented groups, reinforcing workplace inequality. This study presents a computational framework using Natural Language Processing (NLP) and Machine Learning (ML) to detect and analyse such bias. The methodology involves text preprocessing, gender-coded word scoring, topic modelling with Latent Dirichlet Allocation (LDA), clustering via KMeans, and visualisation through t-SNE. A curated lexicon of masculine- and feminine-coded words assigns bias scores, while topic modelling uncovers latent themes in postings. Clustering groups of semantically similar descriptions enables analysis of bias distributions across occupational categories. Findings show that bias varies by job type: technical and managerial roles tend to use more masculine-coded language, while service and support roles favour feminine-coded terms. Semantic cluster visualisations confirm systemic patterns in word usage. This research underscores the need for fairness-aware audits in recruitment, offering both theoretical and practical insights into bias detection. The framework provides organisations with a scalable tool to identify and mitigate hidden biases, promoting inclusive hiring practices and supporting compliance with ethical and regulatory standards.