AN EXPLAINABLE AI DECISION ASSISTANT WITH DOCUMENT GROUNDED QA BIAS WARNINGS AND HUMAN IN THE LOOP CONTROL

Read full text Summarise with VIRAT
Authors: Narsaiah Gallamalla, Koushik Yadagiri, Shashank Peddi, Sainath Mittamedi, Marri Laxman Reddy Institute of Technology and Management
Cited by: 0

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

Project Title: An Explainable AI Decision Assistant with Document-Grounded QA, Bias Warnings, and Human-in-the-Loop Control Overview Modern AI assistants often operate as “black boxes”, producing opaque outputs that may lead to hallucinations, hidden biases, and a decline in user trust.This research introduces a web-based Explainable AI (XAI) Decision Assistant designed to bridge the gap between theoretical XAI principles and practical, user-friendly applications.The system operationalizes transparency and interpretability to support non-technical users in making informed and ethical decisions. Core Features 1. Document-Grounded Question Answering (QA) To minimize AI hallucinations, the system generates responses exclusively based on user-uploaded documents (PDF, DOCX, or TXT).Each answer includes source-cited explanations, ensuring verifiability and factual grounding. 2. Real-Time Bias Detection The system automatically scans queries and reference documents for sensitive attributes — specifically age, gender, and religion — using a rule-based keyword detection mechanism.When detected, the assistant issues prominent warnings to promote ethical AI usage and enhance user awareness. 3. Human-in-the-Loop (HITL) Control Users maintain full oversight through an interactive control interface that enables them to Accept, Reject, or Modify AI-generated responses.This feedback loop ensures continuous accountability and user-driven refinement of outputs. Technical Implementation Backend: Implemented using Python (Flask) for API handling and seamless GenAI model integration. Processing: Utilizes PyPDF2 and python-docx for structured document parsing and indexing. Frontend: Developed with HTML and CSS to deliver a responsive, intuitive user experience. Validation and Testing Informal testing conducted with peers at Marri Laxman Reddy Institute of Technology and Management (MLRITM) evaluated the system across multiple transparency and usability metrics: Bias Warning Usefulness: 4.8 / 5 Overall Transparency: 4.6 / 5 Clarity of Explanations: 4.5 / 5 These results highlight strong user confidence in the system’s interpretability and ethical safeguards. Institutional Context This project was developed as part of the final-year B.Tech program at MLRITM, under the supervision of Assistant Professor G. Narsaiah.It is currently under peer review for the 3rd International Conference on Artificial Intelligence Trends and Pattern Recognition (ICAITPR 2026) with Paper ID: 150.

Fields of Study & Subjects

Explainable Artificial Intelligence (XAI) Ethics and Social Impacts of AI Artificial Intelligence Applications

Publication details

Journal / source
Zenodo (CERN European Organization for Nuclear Research)
Publication date
21 March 2026
Language
EN
Document type
Preprint