Farzaneh Mahdavi

I’m an Agentic AI Engineer who builds agentic LLM and RAG systems that turn unstructured text into useful, structured outcomes. I enjoy moving from proof-of-concept to robust pipelines by combining careful prompting with dependable engineering.

At BMW, I developed proof-of-concept multi-agent RAG systems, working with RAG, LangGraph, LLMs, and AWS. At FIZ Karlsruhe, I built an LLM-based system to extract structured knowledge from materials engineering experimental standards, using LLMs, prompt engineering, ontologies, and Python.

In my UKP Lab work, I investigated open-source and closed-source LLM performance for identifying issues in papers and peer reviews, then developed methods to help reviewers catch missed problems. I also contributed to LiveCoWrite, a collaborative multi-agent writing system—building backend components for real-time writing with LLM-agent support and implementing human-in-the-loop interaction, leveraging LLMs, RAG, agents, Python, Phidata, LangGraph, LangChain, Hugging Face, and SQLite.