From research to working models

You’ll advance the research behind our tabular foundation models. You’ll develop ideas for how models learn from structured data, design experiments to understand why they work, and build stronger pretrained models.

What you’ll do

• Develop model architectures, training objectives, and in-context learning methods for structured data.
• Design synthetic data generation methods and pretraining tasks that help models generalize to real datasets.
• Evaluate predictive quality, robustness, and efficiency across diverse benchmarks.

What we’re looking for

• Experience turning a research idea into a working implementation and evaluating it against meaningful baselines.
• Strong Python skills and experience implementing and training models with a deep learning framework such as PyTorch.
• A solid understanding of deep learning, probability, statistics, and experimental design.
• The ability to read research papers, formulate hypotheses, and design effective experiments.

Nice to have

• Experience researching tabular foundation models, including architectures that model interactions across rows and columns.
• Experience generating tabular data using structural causal models, Bayesian networks, or probabilistic programs.
• Experience running pretraining across multiple GPUs over days or weeks.
• Experience evaluating tabular models under distribution shift, missing data, or changes in feature types.

You’ll build systems that serve our tabular foundation models (TFMs) efficiently. You’ll develop inference infrastructure for diverse datasets and deployment environments, and turn customer problems into improvements to our products and research.

What you’ll do

• Design and improve inference pipelines for pretrained TFMs.
• Extend TFMs to support a wider range of data and query types.
• Deploy models through APIs, on-premises installations, and other formats suited to each customer environment.
• Evaluate predictive quality, inference speed, scalability, reliability, and usability.
• Investigate recurring customer needs and failure cases, and use the findings to improve product capabilities.

What we’re looking for

• Experience developing models with a deep learning framework such as PyTorch.
• Strong Python skills and the ability to write testable, maintainable code.
• A solid understanding of machine learning training, inference pipelines, and data processing.
• The ability to analyze experiments systematically and produce reproducible results.
• The initiative to turn ambiguous problems into clear goals and follow through on a solution.

Nice to have

• Experience with transformers, tabular models, graph neural networks, or in-context learning.
• Experience with distributed training, GPU profiling, or inference optimization.
• Experience building and operating model-serving systems, APIs, or data pipelines.
• Experience deploying machine learning systems in cloud or on-premises environments.
• Experience applying recent research to real-world machine learning systems.

You’ll help define the role Nums AI needs next. You’ll identify an important problem we may not have considered and propose how you would help solve it.

What we’re looking for

We welcome proposals across roles and fields. There are no fixed requirements for title, experience, or field of study.

What to include

• Describe the role you believe Nums AI needs next.
• Explain the problem you would solve and the outcome you would aim to achieve.
• Share the experience or body of work that shows why you would succeed in this role.
• Outline what you would aim to accomplish in your first three months.