Foundation Model Research Engineer Intern

Datarobot; City of Melbourne; On-site; AUD 90,000 - 120,000; Full time

Job summary

DataRobot is seeking a research-engineering intern at post-doctoral level to shape the architecture of a probabilistic foundation model. The role focuses on encoders for temporal, tabular, and mixed-modality data, attention mechanisms, and distributional output heads, with production-quality PyTorch implementation.

The ideal candidate has deep PyTorch expertise, strong foundations in probability and statistics, and a track record of rigorous experimentation and reproducible results.

Qualifications

• Strong PyTorch skills with hands-on experience training deep models.
• Solid understanding of Transformers and long-context architectures.
• Experience with probabilistic modeling and uncertainty-aware learning.
• Strong foundation in probability and statistics.
• Ability to translate architectural ideas into production-quality PyTorch code and experiments.
• Strong engineering habits with readable code and reproducible experiments.
• Ability to debug training instability and iterate quickly.

Responsibilities

• Design core model architecture: encoders for temporal, tabular, and mixed-modality data.
• Run controlled architecture studies and translate results into design decisions.
• Build scalable PyTorch implementations for larger input/output spaces.
• Study how data-generation and inference constraints affect benchmark quality.
• Extend synthetic-data engine with richer stochastic dynamics.
• Turn research ideas into robust implementations with credible results.

Skills

PyTorch

Transformers

Uncertainty-aware learning

Probabilistic modeling

Deep learning

GPU-accelerated training

Education

PhD in ML/CS

Tools

Git

Linux

CUDA

Job description

DataRobot is seeking a research-engineering intern at post-doctoral level to shape the architecture of a probabilistic foundation model. The role focuses on encoders for temporal, tabular, and mixed-modality data, attention mechanisms, and distributional output heads, with production-quality PyTorch implementation.

The ideal candidate has deep PyTorch expertise, strong foundations in probability and statistics, and a track record of rigorous experimentation and reproducible results.