Coursera Flash Sale
40% Off Coursera Plus for 3 Months!
Grab it
Learn to build a comprehensive five-agent system designed to combat digital misinformation throughout its entire lifecycle in this 28-minute conference talk from the Agents in Production Virtual Conference. Discover how to move beyond single-LLM limitations by implementing specialized agents—Classifier, Indexer, Extractor, Corrector, and Verification—that work together to automate fact-checking processes traditionally handled by expert teams. Explore the technical details of each agent, including model sizing and fine-tuning strategies such as using small, fine-tuned encoder models for the Classifier's multi-class labeling versus employing strong reasoning LLMs for the Corrector Agent. Master the construction of an efficient Indexer Agent with hybrid keyword and vector embedding reranking capabilities, understand how the Corrector Agent leverages external search APIs for cross-validation, and learn the role of the Verification Agent as the final quality assurance checkpoint. Gain insights into agent coordination protocols, cost optimization strategies, and comprehensive evaluation methodologies including offline evaluation, online A/B testing, and post-deployment metrics. The presentation draws from ICWSM research and provides a practitioner's guide to building scalable, modular, and explainable systems that can save millions while increasing efficiency through human-in-the-loop approaches.
