LLM/RAG Search & Retrieval

Built from scratch at Fuchs und Eule

Own the retrieval and evaluation stack for an LLM-first search product over 350k+ German/English technical documents (energy regulations, DIN standards, BAFA/KfW funding rules).

Designed query routing, LLM-driven multi-step query decomposition, LangChain-based document chunking (markdown, recursive, semantic), BGE embeddings with cross-encoder reranking over Chroma, and k-tuning.

Built a living query set and MLflow-tracked evaluation workflow covering retrieval precision and answer quality.

Used daily by 40+ consultants and analysts.

• Company: Fuchs und Eule
• Location: Berlin, Germany
• Status: In production
• Relevant Technologies: Python, FastAPI, Pydantic, PyTest, Docker, AWS, LLM/RAG, LangChain, BGE embeddings, Chroma, cross-encoder reranking, MLflow