International Journal of Scientific Research and Engineering Development

Large language models (LLMs) are increasingly deployed in high-stakes settings where factual reliability is essential, yet they remain prone to hallucination — generating fluent statements that are unsupported or contradicted by evidence. This paper presents TruthLens, a multi-agent architecture for real-time evidence verification and hallucination detection. Unlike single-agent retrieval-augmented generation (RAG) pipelines that treat retrieval and generation as a single undifferentiated step, TruthLens decomposes the verification task across five cooperating agents — Orchestrator, Retrieval, Evidence Verification, Hallucination Detection, and Response Synthesis — coordinated through a shared vector memory and an explicit feedback loop. Each agent contributes a specialized, auditable judgment, and claims are only accepted when independently corroborated evidence crosses a confidence threshold. We describe the architecture, agent workflow, coordination algorithms, and a reference implementation built on open frameworks. We evaluate TruthLens against a single-agent RAG baseline and a static factchecking pipeline on a curated benchmark of open-domain and adversarial factual queries, measuring hallucination rate, factual accuracy, task completion rate, latency, and cost. TruthLens reduces the observed hallucination rate by 46.8% relative to the single-agent RAG baseline while incurring a moderate latency overhead, and an ablation study confirms that the Hallucination Detection and Evidence Verification agents contribute the largest share of this improvement. We conclude by discussing failure modes, security and privacy considerations, and directions for adaptive, self-learning multi-agent verification systems.

Jahnavi Somaraju, K. Ramyasree, K. Jashnavi, B. Madhu Sri, Y. Sumathi, "TruthLens: Real-Time Evidence Verification and AI Hallucination Detection for Large Language Models via a Multi-Agent Architecture" International Journal of Scientific Research and Engineering Development, V9(4): Page(480-484) May-June 2026. ISSN: 2581-7175. www.ijsred.com. Published by Scientific and Academic Research Publishing.