Senior LLM Research & Training Engineer

Pragmatike; San Francisco (CA); On-site; USD 200,000 - 350,000; Full time

Job summary

Pragmatike in San Francisco seeks a highly skilled Member of Technical Staff specializing in ML and Large Language Models to advance state-of-the-art foundation models. You will drive research and systems work across training, inference, and optimization on multi-GPU clusters.

You will collaborate closely with researchers and engineers, owning components from theory to production, focusing on performance, scalability, and new model architectures.

Qualifications

• Strong background in ML/LM research or engineering with hands-on experience beyond API usage.
• Experience with large language models, pre-training, fine-tuning, or RLHF is a plus.
• Proven ability to design and optimize training pipelines and infrastructure.
• Familiarity with CUDA/Triton and multi-GPU distributed training is desirable.

Responsibilities

• Research and implement new techniques for training and improving large language models.
• Build and optimize large-scale pre-training and post-training pipelines.
• Improve training throughput, stability, and scalability across GPU clusters.
• Design and optimize model-parallel and data-parallel training strategies.
• Collaborate on software/hardware co-design to maximize training performance.
• Contribute to internal research infrastructure and potential open-source initiatives.

Skills

Python

PyTorch

CUDA

Triton

JAX

Education

Bachelor's degree in a related field

Tools

Distributed training frameworks

Custom kernels

GPU profiling tools

HPC libraries

Job description

Pragmatike in San Francisco seeks a highly skilled Member of Technical Staff specializing in ML and Large Language Models to advance state-of-the-art foundation models. You will drive research and systems work across training, inference, and optimization on multi-GPU clusters.

You will collaborate closely with researchers and engineers, owning components from theory to production, focusing on performance, scalability, and new model architectures.