/benger benchathon 2026
Multi-agent AI · Legal NLP · Python · Top score: 18/18
I built a multi-agent AI pipeline for the BenGER Benchathon 2026, a competition testing whether AI systems can solve German Staatsexamen-level legal cases as well as trained jurists.
The system decomposes complex exam questions into structured sub-problems, routes each one to a specialized agent, and recombines the reasoning into a final answer. It scored 18/18, the highest result in the competition, without any legal training on my side.
The pipeline parses case facts, identifies relevant legal norms, applies structured subsumption logic, and produces a coherent exam-style answer. Each stage runs as an independent agent with its own prompt strategy and tool access.