This paper proposes a conversational agent - chatbot framework that adopts a two-step text similarity approach in order to retrieve relevant answers from large knowledge bases. The proposed "NADINE-bot" is an online question-answer (QA) system able to respond to asylum seekers and other vulnerable migrants' categories about EU related administrative questions, in their native language. The developed agent combines several state-of-the-art models in order to initially retrieve the most relevant document and then, by splitting it into paragraphs, it seeks for an answer in the most relevant paragraph. The NADINE-bot working knowledge source is a collection of administrative tasks, related frequently asked questions (FAQs) concerning EU countries of reception. Nonetheless, for testing purposes the methodology has been applied it to a vast collection of Wikipedia articles, displaying satisfactory results.