Research Scientist, DeepL
Apr, 2022 - Present
Trained, evaluated, and deployed machine translation models — from encoder–decoder architectures to, more recently, LLMs — on top of large-scale internal ML infrastructure (Megatron-based distributed training, evaluation, and inference pipelines).
Built and maintained the backend for serving those models — input preprocessing, output postprocessing, and orchestration of multi-model, multi-step translation pipelines.
Generated synthetic training and evaluation data for a range of MT tasks using heuristics, rules, and LLMs — both in-house models and third-party batch APIs.
Designed and managed human annotation projects for training and evaluation data.
Built data-extraction pipelines to mine MT training and evaluation examples from sources such as arXiv PDFs and LaTeX.
Interviewed Research-team candidates on foundational mathematics.