Generative AI Hallucinations: Is there a practical, *current* fix for internal company data?

We're trying to deploy a custom LLM for internal knowledge retrieval, but it frequently 'hallucinates' information not present in our source documents. We've tried RAG, but the issue persists, especially with nuanced queries. I'm keen to hear if anyone has implemented a robust, real-world solution to significantly reduce hallucinations when dealing with proprietary, factual data. What approaches have actually worked for you?