About the project

The Paritätischer Wohlfahrtsverband Berlin runs Socialmap, a directory of social services across the city. The directory is comprehensive but hard to navigate for the people who need it most: newcomers, people with limited German, and people who need information in Leichte Sprache (Easy Language).

Multilang Socialmap, led by Maximilian Greiner, adds a multilingual RAG-based chatbot on top of Socialmap so anyone can ask questions in their own language and get an answer grounded in real, verified social services data.

The project was a winner of N3XTCODER's Open Innovation Programme and was pitched at re:publica in May 2024.

The challenge

Socialmap holds the answers, but the answers are locked behind a directory interface in German. People in vulnerable situations – often the people who need social services most urgently – cannot navigate it. Translation is not enough; many users need plain-language explanations and culturally appropriate phrasing, not just word-for-word German.

The solution

A retrieval-augmented generation (RAG) chatbot grounded in Socialmap's verified service data. Users ask questions in their own language. The system retrieves the relevant Socialmap entries, then generates a response in the user's language – including Leichte Sprache when requested. Answers cite the underlying Socialmap entries so users can follow up directly with the service provider.

Because the system is grounded in real data, it does not hallucinate services that do not exist – a non-negotiable requirement when people are looking for housing, legal aid or crisis support.

The outcome

• Open Innovation Programme winner, selected from a competitive cohort

• Pitched at re:publica 2024

• Demonstrated multilingual + Leichte Sprache RAG on real social services data

• Proof of Concept that grounded AI can widen access to public services without compromising accuracy