The French Interministerial Digital Directorate's (DINUM) Lab IA team developed NLP-powered search systems to improve information retrieval on government websites, using Haystack as their tool. The team aimed to create a flexible and easy-to-use website search software that could be implemented quickly by government ministries. They built domain-specific question answering pipelines for various government bodies, leveraging the Haystack Pipelines to match their needs. To adapt these pipelines to individual contexts, they created an annotated dataset called PIAF, which is modeled on SQuAD but relies on volunteer annotators. The team designed a framework for automatic parameter tuning called PiafML and implemented a reranking model called Piaf-ranker, which led to an absolute improvement of 12% recall on three different datasets. Their work has the potential to improve data literacy and encourage participation in the political process by providing citizens with access to relevant information through intuitive search systems.