Evidence over claims
Every result is a number on a frozen evaluation set, measured before and after. If a model only reaches parity, we write parity.
Careers · Sevnica, Slovenia · remote-first
Build models that ship.
MediaAtlas is a small engineering team that trains language models on customers' data and deploys them where the data lives. We don't run a hiring funnel. We meet people whose work we can see.
How we work
What working here looks like
Whole problems
You follow a task from raw documents to dataset, training, evaluation and deployment. No hand-offs between departments, because there are none.
Direct contact with customers
The engineers who train a model are the ones on the call with the customer. You hear the problem first-hand and see what your work changes.
Areas
Where we need people
No fixed openings right now. We take open applications in these areas:
- Model training and evaluation: fine-tuning, continued pre-training, benchmark design, PyTorch
- Data and pipelines: dataset engineering, OCR and extraction, RAG and knowledge graphs
- Speech: speech recognition and synthesis, Slovenian first
- Product engineering: .NET and React for our SaaS products Sensoram and EDC
- Language and domain expertise: linguists and domain reviewers (medical, legal) for datasets and evaluation
How to apply
- Send links to work you've done: GitHub, Hugging Face, papers or a product you shipped. They count for more than a CV.
- Add a few sentences on what you'd like to work on and why.
- If there's a fit, we set up a call about the actual work.
Details
- WhereRemote-first, headquartered in Sevnica, Slovenia
- LanguagesEnglish and Slovenian
- Contactinfo@mediaatlas.si, subject "Careers"