Copyright © 2026 Authors retain the copyright of this article. This article is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

@article{199911,
        author = {Pranav Marekar and Aarush Parate and Jeet Ghegade and Arnav Karmankar and Prof Anuja Gaikwad},
        title = {TruthSetu: An AI-Based Multi-Agent System for Real-Time Crisis Misinformation Detection in Environmental Disaster Contexts},
        journal = {International Journal of Innovative Research in Technology},
        year = {2026},
        volume = {12},
        number = {12},
        pages = {1292-1302},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=199911},
        abstract = {Environmental crises — including cyclones, floods, dam failures, and seismic events — generate high-volume misinformation on platforms such as WhatsApp, accelerating public panic and impeding coordinated disaster response. Existing fake news detection systems are predominantly content-centric and fail to operate at the speed, scale, and linguistic diversity demanded by such crisis environments. This paper presents TruthSetu ("Truth Bridge" in Sanskrit), a real-time, multi-agent AI system designed specifically for the Indian crisis information ecosystem. TruthSetu deploys a five-stage pipeline — SCOUT, VERIFY, TRANSLATE, DEPLOY, and LEARN — integrating large language model (LLM) reasoning, Retrieval-Augmented Generation (RAG) over a FAISS semantic index, and automated WhatsApp delivery to return verified corrections to citizens within three to ten seconds of query receipt. The system incorporates trust-weighted source hierarchies privileging government and fact-checking outlets, multilingual response generation in Hindi and Marathi via the Groq LLaMA-3.1-8b-instant model, and a continuous self-improvement loop that seeds verified false claims as retrievable templates. Empirical evaluation of the underlying machine learning detection module using nine classifiers on a 44,898-article dataset demonstrates that XGBoost achieves peak performance with an accuracy of 0.9967 and an F1-score of 0.9964. We further situate TruthSetu within the broader sociotechnical literature on fake news detection, drawing connections to the SHAPE framework (Veerasamy and Badenhorst, 2026) and arguing that effective crisis misinformation detection demands the integration of computational precision, social-network awareness, and human-centred design.},
        keywords = {Fake news detection, Environmental crisis misinformation, Multi-agent systems, Retrieval-Augmented Generation, WhatsApp, Natural language processing, Machine learning, FAISS, Sociotechnical systems},
        month = {May},
        }