Publication

Abstract : Modern recruitment is still a challenge to efficiently process resumes in large numbers. Usually, traditional keyword-based search techniques do not capture the semantic relevance of any skill or experience possessed by the candidate, resulting in bad shortlisting. The Retrieval-Augmented Resume Processing System (RARP) described in this paper is based on document parsing, text chunking, and transformer-based embeddings to improve candidate matching. The system handles 221 different domains by converting resume data into dense vectors for semantic similarity analysis purposes. The system upgrades resume ranking, candidate job matching, and information retrieval for recruiters by integrating retrieval-based techniques into them. Our work demonstrates that the Retrieval Augmented Generation (RAG) system, built with DeepSeek’s LLaMA-8B and using a model all-MiniLM-L6-v2 from Sentence Transformers, improves the resume search and scalability, which is a desirable feature for machine-aided recruitment automation. The system’s effectiveness is supported by cosine similarity results in resume evaluation.