A recruitment support system based on Large Language Models and Retrieval-Augmented Generation

This paper proposes a method for building a recruitment support system that leverages Retrieval-Augmented Generation (RAG) and large language models to automate the processing and analysis of job applications. The system employs a modular architecture that integrates a vector database for semantic retrieval, agent-based query processing, and cloud-based storage and computation to ensure scalability and cost-efficiency. The implementation techniques include extracting data from PDF applications, chunking text into 256-token segments, embedding with Amazon Titan Text Embeddings v2, and generating responses using Anthropic Claude 3 Haiku with a 200,000-token context window. Deployed on AWS, the system optimizes resource usage by employing serverless computing. We evaluated the system using a dataset of 400 test cases, curated from 800 curriculum vitae, utilizing the RAGAS framework. The results demonstrate strong performance, achieving a Context Recall of 0.9183, Faithfulness of 0.882, Factual Correctness of 0.753, and Semantic Similarity of 0.827. A user-friendly interface supports semantic searches, keyword queries, and direct CV interactions, which enhances practical applicability. The system effectively streamlines candidate screening, with high retrieval accuracy and response fidelity.