Rethinking Retrieval-Augmented-Generation for a Citizen Services Digital Assistant

The current retrieval-augmented generation technology does not guarantee factual correctness, and the interface may encourage users to rely on the generated text instead of double-checking with the cited sources. This project focuses on making retrieval-augmented generation reliable enough to be used in high-risk information access applications for the public sector. The goal is to create a better blend of 'helpful' responses coming from large language models, and reliability of traditional search, ensuring that the users receive the best information available, while considering both the time spent and risk of automation bias.