VERA: A WHATSAPP-BASED RETRIEVAL-AUGMENTED GENERATION SYSTEM FOR AUTOMATED HEALTH MISINFORMATION FACT-CHECKING IN LOW- AND MIDDLE-INCOME COUNTRIES

Abstract

Health misinformation is a growing public health emergency, with disproportionate impact in low- and middle-income countries (LMICs), where fragile information ecosystems increase susceptibility to infodemic harm. This paper presents VERA (Verified Evidence Retrieval Assistant), a WhatsApp-based fact-checking system that combines Retrieval-Augmented Generation (RAG) with an open-source large language model (LLM) to counter health misinformation in resource-constrained settings. VERA accepts user-submitted health claims, forwarded messages, or article links through WhatsApp, retrieves evidence from PubMed, Semantic Scholar, and WHO/CDC sources, and returns a structured verdict, supported, refuted, nuanced, or evidence-insufficient, with citations and a plain-language explanation. Llama 3 8B serves as the reasoning backbone, chosen for its competitive biomedical performance under RAG, zero licensing cost, and deployment feasibility on modest cloud infrastructure. VERA is a theoretical architecture and hasn’t been deployed or empirically validated therefore, the contribution of this paper is the design itself, together with a rigorous evaluation protocol, precedent from comparable deployed systems, and worked examples demonstrating how the pipeline would process representative claim types. We detail the complete technical architecture, the rationale behind each design choice, a multi-dimensional evaluation framework spanning technical accuracy, usability, and public health impact, and the governance safeguards required before real-world deployment. VERA demonstrates that scientifically rigorous, evidence-grounded fact-checking can be delivered through infrastructure LMIC populations already use and can afford.