Research Scientist, Pretraining and Scaling (PhD)

About Ataraxis AI

Ataraxis AI is a VC-backed start-up working at the intersection of multi-modal AI, oncology, and precision medicine. We are building cutting-edge diagnostic tools to assist physicians in selecting the most optimal treatments for their patients.

We are a team of AI researchers, medical doctors and operators advised by a team of advisors who are pioneers in their respective fields, including the AI godfather, Yann LeCun, and oncologists at top cancer research institutes.

At Ataraxis, you will have a unique opportunity to shape not just the future of our company, but the future of healthcare. We are looking to bring on highly motivated candidates to fulfill this mission.

We operate with a flat organizational structure where every team member is expected to actively contribute. Leadership roles are earned by those who demonstrate initiative and deliver consistently high performance. Strong work ethic and the ability to prioritize effectively are essential. All candidates must communicate clearly, sharing insights accurately and concisely with teammates. We don’t use recruiters — each application is reviewed personally by a relevant team member and/or founders.

Responsibilities

• Design and implement novel self-supervised learning approaches for gigapixel pathology images and other biomedical data modalities.

• Rapidly implementing the latest state-of-the-art methods from the deep learning literature.

• Build scalable training pipelines, including profiling, debugging, and optimizing distributed GPU training.

Qualifications

• PhD in machine learning or statistics.

• Machine learning expertise with a deep understanding of scaling large models.

• Deep understanding of model training through careful ablations.

• Familiar with distributed training and parallelism strategies. Experience building and scaling training in HPC clusters.

• Passion for research. Strong preference will be given to candidates with published papers in A* conferences (ICML, ICLR, NeurIPS).