Inference & Model Quality Engineer

Role overview Help run open language models efficiently on GPU infrastructure and establish a consistent way to measure their quality. The work spans model serving, performance and cost tuning, evaluation, and improvements to an in-house Dutch model, with results documented for users and regulatory transparency.

Responsibilities - Deploy and operate models on GPUs using serving frameworks such as vLLM, with SGLang planned. - Assess hardware fit and serving cost through VRAM estimates, quantization, and related trade-offs. - Benchmark throughput and latency, then improve performance and cost per token. - Build a shared evaluation harness covering quality, Dutch-language ability, grounding, tool use, and speed. - Create datasets, run fine-tuning experiments, and compare new model versions against prior ones. - Publish model cards and benchmarks, including required AI Act transparency information.

Requirements - Experience serving models on GPUs, fine-tuning language models, or both. - Familiarity with approaches such as SFT, DPO, or LoRA, or model-serving tools such as vLLM and SGLang. - A measurement-led approach using benchmarks, evaluations, and statistics where appropriate. - Python, Linux, containers, and common ML tools such as Hugging Face or PyTorch. - Fluent English, residence in the Netherlands, and the ability to work independently.

Nice to have - Dutch language skills, especially for judging the quality of Dutch model outputs.

Benefits and work setup - Remote work, with a headquarters in Groningen. - Competitive salary, described as top of market. - Team weeks abroad twice a year and an early role with meaningful ownership.