Reliable LLM backends with Python and FastAPI: Dishant, Berlin

Hi everyone, I'm Dishant, an independent AI engineer based in Berlin. I focus on backend systems and LLM applications, mostly in Python with FastAPI, Django, PostgreSQL and Docker. My day to day tools include Claude, Cursor, LangGraph and Claude Code.

Right now I'm building production-style projects around AI workflows. One is a FastAPI backend for AI-assisted job analysis, with an LLM provider abstraction, structured outputs, background jobs and CI. Another is a local-first app that reads Gmail job application emails, classifies status updates and exports them to CSV. I care most about making AI systems reliable and correct, so retrieval and evaluation are where I'm spending my time.

I'd love an intro to anyone who has put LLM evaluation into a real production workflow, or feedback on how you test RAG systems beyond spot checks.