FAVA, a retrieval-augmented language model, is designed to detect and correct hallucinations in AI outputs, which can lead to serious consequences in fields like healthcare, education, and journalism. The approach combines evidence retrieval with fine-grained error detection and correction, using a taxonomy of six distinct types of hallucinations. FAVA's training on diverse synthetic data enables it to generalize well to unseen errors and outperform its counterparts across various hallucination types. While limitations exist, such as reliance on external sources for evidence and challenges with complex claims, future improvements in retrieval processes, data generation, and taxonomy expansion will further strengthen FAVA's capabilities. As AI continues to shape industries, tools like FAVA are crucial for ensuring reliability and trust in AI systems.