Rezumy parses an uploaded resume (PDF or DOCX), cleans and segments the text with spaCy/NLTK, then generates 768-dimensional semantic embeddings via SentenceTransformers/BERT. Multiple scoring modules (ATS, skills, experience, projects, education) evaluate the resume, cosine similarity matches it against a target job role, and Gemini 2.5 generates personalized, actionable rewrite suggestions — down to specific course recommendations for each skill gap.