Senior Engineer – Distributed Training Systems & Performance

Senior Engineer – Distributed Training Systems & Performance

Squeeze every FLOP out of a very large cluster. Sharding, numerics, checkpointing, MFU.

Location: Delhi NCR or Bengaluru | Experience: 4 – 12 years

About the opportunity

We are retained by a well-funded AI lab building foundation models from the ground up in India — real compute, real pretraining, not fine-tuning on top of someone else's model. The team is small and the work is unusually direct: what you build ships into the model.

The client's identity is confidential at this stage and will be shared on first call.

What you will do

• Optimise core training infrastructure across the JAX/XLA stack.

• Design tensor, pipeline and data sharding strategies across large TPU/GPU clusters.

• Implement FP8 numerics and architect reliable high-speed checkpointing.

• Profile workloads continuously to maximise Model FLOPs Utilisation.

What we are looking for

• 4+ years in distributed systems, HPC, or ML infrastructure.

• Deep expertise in parallel computing and hardware accelerators.

• Low-level stack optimisation experience.

What is on offer

• Compensation is open and benchmarked to the top of the Indian market for this profile, with meaningful equity.

•  Delhi NCR is the primary base, but the client is flexible on Bengaluru for the right person.

• Relocation support, including for candidates returning from outside India.

How to apply

Apply here, or write to [Confidential Information] with your CV and links to anything you have published, shipped or open-sourced. We reply to every serious application within 48 hours.