Research Engineer, Foundation Models

Brahma Consulting Group; San Francisco (CA); On-site; USD 150,000 - 210,000; Full time

Job summary

Brahma Consulting Group in San Francisco seeks a Research Engineer for Foundation Models who will train models from scratch and own pretraining work end-to-end.

The role emphasizes architecture decisions, data pipelines, and running training runs, using PyTorch and JAX in a distributed multi-node setting.

This early-stage lab values hands-on execution and projects with practical transfer to physical systems.

Qualifications

• Experience training models from scratch, not only fine-tuning.
• Ownership of end-to-end training runs and debugging training runs.
• Hands-on work with data infrastructure enabling large-scale runs.

Responsibilities

• Pretrain foundation models for physical systems with distributed multi-node setups.
• Make architecture decisions, design data pipelines, and run training runs.
• Contribute to a small, fast-moving team focused on impactful results.

Skills

End-to-end training

Distributed training

Data pipeline development

Tools

PyTorch

JAX

Job description

Research Engineer, Foundation Models. Early-stage lab in SF, well funded, small team. You'd train models from scratch rather than adapt existing ones.

Client is building foundation models for physical systems. The team is small enough that you'd own real surface area, and the work is pretraining: architecture decisions, data pipeline, and the training runs themselves.

What we care about:

You've been close to model training, not just model usage. Maybe you've owned a run end to end, maybe you've been the person debugging a loss curve at 2am, maybe you've built the data infrastructure that made a run possible. We're more interested in what you've touched than in which frameworks you touched it with.

Stack is PyTorch and JAX, distributed training across multi-node setups. If you've worked in that world, great. If you've done equivalent work in a different stack, we'd still like to talk.

Worth knowing: this is a pretraining role. If your experience is primarily fine-tuning or adapting released models, it's probably not the right fit, though we'd still encourage you to apply if you think the underlying skills transfer.

Skills tags

Broader than the req warrants, on purpose. These feed LinkedIn's recommendation engine more than they feed candidate screening.