Case study · AI implementation
AI assistant (RAG) for a large public institution
The assistant answers from the institution’s documents and links to its sources. Accepted by the client.
01 · Problem
A large public institution needed an assistant that answers questions from its own documents. Requirements: data in the EU, login through the existing SSO and no personal data in the index.
02 · Solution
A RAG architecture on the Polish Bielik model, with search in Qdrant (bge-m3 / e5 embeddings) and Whisper for transcribing recordings. Before indexing, documents are anonymised with Presidio and a Polish SpaCy model (PESEL numbers, names and other personal data). Workflows in n8n, the application in Symfony, login via Entra ID / SAML SSO, servers in the EU.
03 · Result
Accepted by the client. Every answer links to the source documents.
AnonymisationClient data anonymised. Company name, numbers and screenshots are published only with consent.
A similar project on your side?
Book a call