INTEGRA: INclusive Technology for Enhanced Gradation and Review of Applicants

Francesco Pivi, Michele Lombardi, Roberta Calegari, Elisa Rimondi

Keywords: Ethics, AI, RAG, GDPR

Abstract:

A company's real strength comes from embracing inclusivity and diversity among its people. Addressing potential biases in resume screening is crucial to enhancing social and workplace well-being. We propose, in collaboration with the company SACMI, a solution called INTEGRA: Inclusive Technology for Enhanced Gradation and Review of Applicants. INTEGRA employs a Retrieval-Augmented Generation (RAG) Fusion architecture to process resumes, effectively mitigating biases before storing data. It includes a bias filter for queries to prevent their introduced by users, intentional or unintentional. Leveraging open-weight large language models (LLMs), INTEGRA makes a significant step towards compliance with the AI Act and General Data Protection and Regulation (GDPR) law in EU, while maintaining data privacy within the organization, ensuring transparency and regulatory adherence.
The evaluations were conducted using the cited Kaggle dataset and in order to assess the effectiveness of the retrieval component we developed, in collaboration with the HR department, a custom question dataset. This dataset will be publicly available, in the appendix, to support reproducibility and encourage further advancements. Our goal with this initiative is to promote fairer HR practices, fostering greater inclusivity and diversity within organizations.

Submission Number: 2