TM
Timur Mikhailov
@durden
AI/ML Engineer building LLM agents and evals
I build LLM systems and the evals that show they work. At Langfuse I built the LLM-as-a-judge pipeline that scores every judge against human labels.
About
At Hardware I fine-tuned NVIDIA NeMo ASR to −5.60 WER points against base and served gemma on vLLM in k3s. At Obscura I built an agent system with six LLM pipelines and a CI gate for evals.