Start a project

Custom models / Applied intelligence / Athens · Hybrid / 5+ years

Senior LLM Training & Fine-Tuning Engineer

Turn domain knowledge into model capability. Own the data, weights and evaluation behind custom models for our customers.

Apply for this role All roles
Designers working at Apple displays in the bright Designco studio

The opportunity

Intelligence. Built for the brief.

We want intelligence that earns its place in a product. You’ll adapt open-weight models on our NVIDIA DGX Spark clusters, build rigorous evaluations and work with our engineers to put the results into customers’ hands. You should be as comfortable explaining a failed experiment as delivering a strong one.

DGX SparkPyTorchHugging FaceNVIDIA NeMo

01 / The mandate

What you’ll own.

  • Build repeatable dataset curation, cleaning, labelling and versioning pipelines around each customer’s domain and permitted data.
  • Own supervised fine-tuning and parameter-efficient adaptation, including LoRA and QLoRA, with reproducible training runs and checkpoints.
  • Plan memory, compute and distributed workloads for our DGX Spark environment. Choose methods that fit the hardware and the problem.
  • Design task-specific evaluations, held-out tests and regression checks. Compare adapted weights against strong baselines before claiming an improvement.
  • Partner with product and engineering on inference, quantisation, latency, deployment and a documented customer handover.

02 / The standard

What you bring.

  • 5+ years in applied machine learning or ML engineering, with hands-on experience training and adapting model weights.
  • Strong Python and PyTorch, practical experience with Hugging Face or NVIDIA NeMo, and the ability to debug a training pipeline beyond the notebook.
  • A shipped fine-tuning project you can explain from dataset and objective through evaluation and deployment.
  • Working knowledge of GPU memory, mixed precision, checkpoints, Linux, containers and multi-node training constraints.
  • Sound judgement about dataset permissions, leakage, sensitive information, model limitations and reproducibility.

03 / The extra edge

Even better.

  • Experience with NVIDIA systems, ARM-based GPU environments, domain-specific or multilingual models.
  • Preference optimisation, synthetic-data quality control, experiment tracking or model-serving infrastructure.

04 / The studio

Close to the work.

Athens-based, with a hybrid rhythm. You’ll work directly with designers, engineers and specialists, with access to the people making the decisions.

Bring judgement, curiosity and a high standard for your own work. We’ll bring ambitious briefs, a multidisciplinary studio and room for your contribution to matter.

First, the person.

Tell us a little about yourself. All fields are required unless marked optional.

Let the work talk.

A public or shareable link. A redacted case study is welcome.

80–5,000 characters. Share only work you have permission to share. Redacted examples are welcome.

A few final details.

Your CV and timing complete the picture.

PDF or DOCX · Maximum 3 MB

We’ll review your application against this role. If there’s a fit, we’ll get in touch to arrange a conversation and a deeper look at your work.

Different strengths. Shared standards.

Other open roles.