Is Fren reliable? How does it avoid AI "hallucinations"?

Fren is designed to minimise the hallucinations typical of generative AI models as far as possible:

• Answers are generated from official sources retrieved in real time (RAG technique: Retrieval Augmented Generation)
• Every statement is accompanied by the exact citation to the specific session, vote or official gazette
• The system recognises when it does not have enough information and states so explicitly instead of making things up
• Constant internal audits of answer accuracy