Enhancing Resume Screening Efficiency and Quality with Retrieval Augmented Generation

Abstract

While resume screening is a crucial stage for any recruitment process, it is also a challenging one. Nevertheless, most traditional screening methods cannot assist recruiters effectively because they are labour-intensive and highly prone to human biases. Moreover, many existing automated solutions are not able to address the complex, context-heavy, and dynamic nature of resumes written in natural language. This presents a research gap for approaches that can resolve the issues inherent in existing screening methods.

To address this gap, this paper presents the implementation and evaluation of a proof-of-concept (POC) Large Language Model (LLM) agent. The primary objective of this system is to assist hiring managers in matching job descriptions with suitable resumes through a query-response mechanism. In practice, recruiters input a job description as the query into the agent so that it searches for the most fitting resumes to produce relevant responses such as analyses or summaries of resumes. To achieve this, the system utilizes Retrieval Augmented Generation (RAG) to integrate the applicant profiles into the LLM’s knowledge base. This technique can significantly enhance the accuracy and relevance of responses of the LLM agent to recruiters’ queries.

For evaluation, the proposed model was assigned the task of matching 500 job descriptions to a suitable resume among a large heterogeneous database of 1000 synthetic applicant profiles. It was discovered that the proposed model showed promising results in terms of resume selection and answer quality. This demonstrates the great potential of the RAG-based system in resume screening and highlights the need for additional research on similar models in this domain.