AI-Based News Verification System Using Large Language Models and Retrieval-Augmented Generation

The pace with which digital news media and social media started proliferating has intensified the rate of misinformation, bringing significant problems with the reliability of information and the credibility of the people. Traditional fake news detection systems rely on the traditional way of machine learning systems, which are fed by predefined data sets, which restricts its flexibility to new events and real-time changes. Also, stand-alone large language models (LLMs) are likely to be susceptible to failing to base their responses on the existing evidence, which in turn leads to a high risk of hallucination and contextual bias. The aim of the paper is to suggest a real-time AI-based News Verification System that will be built by incorporating Retrieval-Augmented Generation (RAG) with a Large Language Model (Gemini 2.0 Flash) to achieve context-sensitive and explainable news content verification. It is founded on a modular architecture of a rest-based architecture written in FastAPI as a backend and Next.js as a frontend, MongoDB as a persistence layer and JWT as an authentication. The Tavily Search API retrieves real-time contextual evidence and then uses it together with the logic of LLM to ensure they become more credible and less groundless. The framework generates ordered output in terms of classification label (Real/Fake), credibility score (0-100%), summary of explanation and identification of suspicious phrases. Performance assessment identifies a mean of 2.8 seconds response latency time when the system is stable with simultaneous API requests. The suggested architecture offers an architecture which offers scalability, modularity, and production readiness to detect misinformation in real-time in dynamic digital environments.