Fullstack Software Engineer at BRB Business IT-Solutions GmbH, based in Hamburg.
I build AI systems in Java. My focus is LLM integration and multi-agent pipelines. Most of my time goes into Faktum, a real-time fact-checking platform for live broadcasts, built on a custom Mistral client and a multi-stage agent pipeline. Alongside that I work on backend systems, concurrency, and developer tooling.
skittles Research framework where LLM agents from different providers trade on a double-auction market, used to observe how each model reasons and strategizes under identical conditions.
ecotopia Political eco-simulation driven by fine-tuned Mistral 8B and 22B models for promise extraction and agent behavior. Built at the MistralAI Hackathon in London.
GuardrailQuestionmark AI-safety research framework exploring LLM decision-making under existential-threat conditions.
agent-docs A Claude Code skill that writes self-staleness-aware documentation briefs for any feature in a repository.
orderbook Thread-safe order book with efficient limit and market order matching and event handling.
avenue Lightweight TCP pub-sub server built on Java virtual threads.
tamboui-spring Spring Boot integration for TamboUI: terminal user interfaces via annotations and dependency injection.
Java 21 and Spring Boot at the core. TypeScript and React on the frontend, Python for ML work. PostgreSQL with pgvector, Docker, and the Mistral and MLX toolchains for everything LLM.