Social engineering remains one of the most pervasive threats in cybersecurity, exploiting human weaknesses rather than technical vulnerabilities. Its impact is amplified by the growing use of diverse communication channels and the increasing realism of AI-generated content. This thesis presents the design and implementation of a lightweight browser extension aimed at detecting potential social engineering attempts directly on the user’s workstation. Unlike traditional approaches, the system focuses on analyzing different types of files and content sources—such as text documents, images, and audio recordings—by integrating Natural Language Processing (NLP), Optical Character Recognition (OCR), and Speech-to-Text (STT) technologies. To validate the proposed approach, different machine learning techniques for text classification are explored and compared.