Auditing Fairness in RAG-Augmented LLM Interview Systems: A Study of Ollama-Based Local Deployment for Automated Technical Hiring

Read full text Summarise with VIRAT
Authors: Y. Chiranjeevi Ashok Kumar, K.Sai Sri Hruthi
Cited by: 0
References: 5

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

The growing adoption of Large Language Models (LLMs) in automated technical hiring systems raises critical concerns regarding fairness, transparency, and evaluation stability. While Retrieval-Augmented Generation (RAG) enhances contextual grounding by incorporating resume-specific knowledge, bias auditing in locally deployed LLM infrastructures remains largely underexplored. This paper presents a fairness auditing framework implemented within a containerized architecture using Ollama running Llama3, orchestrated through Docker-based microservices. The system integrates a Next.js frontend, a Node.js backend for interview orchestration, PostgreSQL-based semantic resume retrieval, and Redis for stateful multi-turn memory management, forming a privacy-preserving RAG-enabled technical linter view platform. To systematically evaluate bias, we design controlled resume perturbation experiments that isolate demographic indicators, institution a background, and linguistic fluency while preserving equivalent technical skills. The system generates adaptive interview questions and structured performance scores across multiple simulated sessions. Fairness is quantified using score variance analysis, difficulty drift measurement, sentiment polarity comparison, and multi-run stability metrics. Statistical comparisons are performed to detect evaluation inconsistencies under controlled variable shifts. To mitigate observed disparities, prompt-neutralization strategies and rubric-based scoring calibration mechanisms are introduced and re-evaluated experimentally. Results demonstrate measurable bias patterns and corresponding reduction trends after mitigation, contributing a reproducible applied AI engineering framework for auditing and improving fairness in RAG-based, locally deployed LLM interview systems.

Fields of Study & Subjects

Ethics and Social Impacts of AI AI and HR Technologies Names Identity and Discrimination Research

Publication details

Journal / source
Research Digest on Engineering Management and Social Innovations
Publication date
31 March 2026
Language
EN