About Synaptiq
Synaptiq was started by cancer researchers and radiation oncologists who got tired of drawing organ contours by hand. We were the first to ship AI tumor segmentation to the clinic. We built an AI to enhance and personalise radiotherapy. It’s CE-marked, segments 250+ structures across CT and MRI, and is deployed in more than 50 clinics.
We use AI tools in our daily work and our interviews. We care about judgment, depth, and research taste, not memorization.
The Role
You’ll own one or two research directions end-to-end, from hypothesis to experiments to clinical deployment. Right now we’re working on 3D self-supervised pre-training, few-shot segmentation, and a multimodal foundation model (vision + language across CT, MR, and clinical reports). But these shift as new tech and results come in, and you’ll help decide where.
This is not a support role. You’ll shape what gets built next, then own the execution; your work ships to the clinic, not to a paper queue.
What You’ll Do
- Lead your research direction(s): design experiments, run them, see what breaks, try again.
- Train and test models on our H200 cluster (PyTorch, Hydra, DDP, mixed precision). When results are ready for the clinic, you work with the oncologists to get them there.
- Leave a clean trail, ablations and results should be reproducible.
- Conferences, patents, clinical studies. Your work gets out there.
- Push back when a direction isn’t working. We’d rather kill a project early than keep it on life support.
- Help build the research team and mentor new hires as we grow.
What We’re Looking For
Must-have
- PhD or equivalent research experience in ML, computer vision, or medical imaging
- Experience completing full research cycles: hypothesis through experiment to documented results (publications, thesis, or technical reports)
- PyTorch expertise: comfortable building custom architectures and attention layers, not just fine-tuning someone else’s repo
- Can own a research direction: frame problems, design experiments, interpret results, course-correct on your own
- Strong experimental workflow: clean ablations, proper baselines, reproducibility discipline
Preferred
- Experience with 3D medical imaging or volumetric data (CT, MRI)
- Familiarity with foundation models, self-supervised pretraining, few-shot learning, or vision-language architectures
- Disciplined experiments, including version-controlled code, documented decisions and reproducible results
- Multi-GPU training (DDP, mixed precision, large batches)
Nice to Have
- 3D vision or segmentation experience from adjacent domains (autonomous driving, remote sensing, robotics)
- Publications at relevant conferences or journals, or projects/challenges (eg MICCAI)
- Experience with experiment management tools (Hydra, W&B, whatever works)
- Comfort with AI-assisted development workflows
What We Offer
- Research ownership: Lead a research direction, not a sub-task. Your name is on the work.
- Compute access: H200 GPU cluster. No queue, no approval needed to run experiments.
- Clinical impact: Every segmentation improvement directly reduces radiation dose to healthy tissue in cancer patients.
- Conference & publication: ESTRO annually, ASTRO, and ML venues. We publish, with obvious IP exceptions. Patents and clinical studies too.
- Research freedom: Core directions first. Beyond that, explore what interests you.
- AI-native culture: Claude, Cursor, and whatever’s best this month.
- Remote. Full-time preferred, but part-time can work, we’ll figure it out.
- Small team, big problems: The decisions you make here would require three levels of approval elsewhere.
Apply by sending your CV and a link to your strongest first-author paper (or the project you can defend most deeply) to office@synaptiq.io. If we move forward, expect a single technical interview (AI tools welcome) focused on how you think through engineering problems.
We’re not looking for someone who memorised the latest survey. We want someone who’s run the experiments that didn’t work, argued with reviewers, and knows why their second-choice architecture was wrong.
Start: As soon as possible
Location: HQ Cluj-Napoca, Romania, but you choose where you work
