How can we guarantee that AI-generated answers always align with Swiss federal law?
Problem
Our in-house OpenWebUI instance already uses a hybrid RAG pipeline (BM25 + embeddings, re-ranked with CrossEncoder and tunable relevance thresholds) to query internal documents. While this yields solid results, it is not yet robust enough for the scale and structure of Swiss legislation. A richer knowledge representation and stronger grounding are needed to avoid hallucinations and improve trust.
Objective
Provide a solution that delivers reliable, explainable answers to legal questions based on Swiss legislation. For example, you could achieve this using one of the following approaches:
- GraphRAG: generate a knowledge graph from Swiss law (RDF is provided) and apply GraphRAG for retrieval + reasoning, or
- Law Agent: implement an agent that answers questions on Swiss law, always returning an accuracy/confidence score and the underlying citations.
Support for Hackers
- Full Swiss federal legislation available as SPARQL endpoint or directly as HTML assets (Metadata, Assets)
- FOITT AI team on-site friday and saturday) plus virtual experts on demand on Saturday.
Technical Preferences
- Prefer solutions that plug directly into OpenWebUI via its RAG function or an external pipeline with a clear REST/GQL contract.
- Use open-source tooling where possible; licensing must allow federal usage.
Why hack? Real-World Impact – Your prototype can land directly inside the Swiss Federal Administration and a premium Dataset – Full corpus of Swiss law.
About the Challenge Partner
The Federal Office of Information Technology, Systems and Telecommunication (BIT/FOITT) is the Swiss government’s in-house tech powerhouse. We run >50 000 workstations, 1 000+ specialist applications and several state-of-the-art data-centres—plus a sovereign GPU stack dedicated to public-sector AI. Your prototype won’t disappear into the void: if it works, it can be deployed on infrastructure that already serves the entire federal administration.