Inference & Model Quality Engineer

Role overview Ensure that language models hosted on GPU infrastructure run efficiently and reliably, while establishing consistent ways to measure their quality. The work spans inference performance, model evaluation and improvement of a Dutch language model, with findings documented through benchmarks and model information.

Responsibilities - Deploy and maintain models on GPUs, using inference engines such as vLLM or SGLang. - Assess hardware fit and operating cost, including VRAM needs and quantization choices. - Benchmark and improve throughput, latency and cost per token. - Build a shared evaluation framework covering model quality, Dutch language performance, grounding, tool use and speed. - Create datasets, run fine-tuning experiments and compare new model versions against prior releases. - Publish model cards and benchmarks, including required model transparency information.

Requirements - Experience running models on GPUs, fine-tuning language models, or both. - Familiarity with evaluation methods, benchmarking and using statistics to assess results. - Python, Linux and container experience, plus relevant tools such as Hugging Face and PyTorch. - Fluent English, ability to work independently and learn quickly. - Current residence in the Netherlands.

Nice to have - Dutch language ability, especially for evaluating the quality of Dutch model outputs. - Familiarity with techniques such as SFT, DPO or LoRA.

Benefits and work setup - Remote work, with a headquarters in Groningen. - Salary described as competitive and at the top of the market. - Twice-yearly team weeks abroad. - Early-team responsibility and influence over the work.