Fake news, misinformation, and disinformation pose serious threats to society by eroding public trust, deepening divisions, and causing harm across various domains. Several existing deep learning-based fact-checking models are limited by their reliance on static assessments of truthfulness, often evaluating information in isolation or assuming that relevant evidence is readily available. These approaches often overlook the dynamic nature of false information, which spreads rapidly across platforms, making it increasingly difficult to trace its origin and verify its accuracy. In this work, we propose MulitWebFacts, a comprehensive framework for fact-checking that integrates information from multiple online sources. It addresses all essential sub-tasks, including check-worthy claim detection, source collection, sentence relevance classification, veracity estimation, and evidence highlighting. In particular, for veracity estimation, we propose a hybrid model that utilizes a token-level augmented Long Short-Term Memory (sLSTM) network to learn the textual content of a claim and combines sentence-level sLSTMs enhanced with hierarchical attention mechanisms to capture the evolving context of information surrounding the claim. Additionally, attention-based evidence highlighting is incorporated to provide insights into sources and contextual factors that contribute most to the verification process, enhancing transparency and fostering trust in results.