Inference & Model Quality Engineer

HostYourAI · Groningen

We run open models on GPUs across Europe and are building our own Dutch model, Loes. You make sure those models run fast, reliably and affordably, and that we know how good each model really is. What you'll do Get models running, and keep them running, on our GPUs: vLLM now, SGLang next. Figure out what fits where and at what price, from VRAM estimates to quantization. Measure and improve throughput and latency, and keep the cost per token sharp. Build an eval harness that holds every model we host to the same standard: quality, Dutch, grounding, tool calling, speed. Make Loes better: build datasets, run finetunes and measure whether a new version really beats the previous one. Publish model cards and benchmarks, including the transparency information the AI Act requires per model. What we ask Experience running models on GPUs (vLLM, SGLang or similar), training or finetuning language models (SFT, DPO, LoRA), or both. You measure instead of trusting your gut: benchmarks, evals, and statistics where needed. Python, Linux and containers, and the usual tooling around them (Hugging Face, PyTorch). You don't need to know everything, learning fast is what counts. Fluent English. Dutch is a genuine plus, because you judge the quality of Loes most sharply in your own language. You must be based in the Netherlands. Independent. There's no manager spelling out your work for you. What we offer A competitive salary at the top of the market. Remote work, with our HQ in Groningen. Twice a year the whole team spends a week abroad together. A spot on the first team of a startup, with all the room and responsibility that comes with it. And above all: the chance to be part of this journey from the very start.

More roles at HostYourAI