Case Study: Trade Republic migrates live LLM workflows to agentic harnesses with Langfuse

Trade Republic migrates live LLM workflows with Langfuse in 2 years

Trade Republic, one of Europe's largest fintechs and a fully licensed bank, faced the challenge of migrating its live production LLM workflows from predetermined prompt chains to more modern and flexible ReAct agent architectures. This migration was considered too risky for a naive rewrite, so they needed a safe method to implement the new agentic setups. They accomplished this using the self-hosted, open-source version of Langfuse.

The solution implemented with Langfuse involved a rigorous evaluation flywheel. Langfuse was used to trace every run, score good and bad examples, and store cases in datasets. For the architecture migration, they utilized homework and exam datasets to have their old workflow and new ReAct agent compete, with an LLM judge comparing the results. This process ensured the new agent performed reliably before being shipped. Using Langfuse allowed the team to successfully navigate strict bank compliance requirements and safely upgrade their LLM infrastructure.

Trade Republic