Senior Software Engineer, AI Runtime

P-1428

At Databricks, we run the world's largest AI platform. We believe the next breakthroughs will come from models trained on domain-specific enterprise data, not from bigger general-purpose models alone.

That's why we built AI Runtime, a serverless product that makes GPU training of any size feel simple, so companies can train their own models the way frontier labs do. Customers across industries already use it to train models on governed data, from automotive AI at Rivian to Merck's TEDDY foundation models for single-cell biology and drug discovery, all the way to agents like KARL from Databricks AI Research.

It takes a lot of engineering to make that work, and we've written about it on our blog: how we keep GPUs reliable, make training fault-tolerant, and speed up fine-tuning with fused kernels and smart sequence packing. As a Senior Software Engineer, you'll work on the systems behind all of this, solving hard technical problems and helping shape how AI Runtime evolves.

The impact you will have:

• Build core parts of AI Runtime and contribute to Databricks' technical direction in AI training infrastructure.
• Develop reliability and observability features for a fleet that keeps growing in scale and complexity.
• Improve AI Runtime's developer experience, so the largest training jobs feel as simple as a first experiment.
• Help prepare AI Runtime for new GPU generations and workload types.
• Take on hard multi-node training problems, from orchestration and GPU scheduling to data loading and checkpointing.
• Make training faster and cheaper by raising GPU efficiency across model architectures and hardware generations.
• Work with research, product, and customers on what we build next and own features from design through production.
• Champion engineering excellence, raise the bar, and help the engineers around you grow.

What we look for:

• 5+ years in large-scale distributed systems, with experience in GPU training, high-performance computing, or ML systems.
• Experience training large models with FSDP, DeepSpeed, or Megatron, and familiarity with parallelism strategies.
• Working knowledge of GPU performance, including accelerator architecture, high-speed interconnects, and kernels.
• Strong understanding of training resilience patterns, such as checkpointing, failure detection, and automatic recovery.
• A background in building and running multi-tenant cloud platforms as products.
• A track record of owning complex features end to end and seeing them through to customer impact.
• The clarity to explain technical designs in writing and in person, and a reputation as a teammate others learn from.
• BS in Computer Science or a related field (MS or PhD preferred).