About the role
At Sarvam AI, the ML Engineer (Training Infra) will be part of the Models team, focusing on developing and maintaining the infrastructure for training large foundational AI models. The role involves working with distributed GPU clusters, optimizing training throughput, and ensuring system reliability for long training runs. The engineer will collaborate closely with research teams to implement efficient parallelism strategies and GPU kernel development, contributing to India’s sovereign AI platform.
The daily responsibilities include building scalable training systems, profiling and tuning GPU workloads, and developing custom CUDA or Triton kernels to enhance performance. The engineer must have experience with large-scale distributed training frameworks like Megatron-LM or DeepSpeed, and possess deep knowledge of GPU architecture and PyTorch internals. The role demands proactive problem-solving to prevent costly training failures and ensure smooth model development.
This full-time, on-site position in Bengaluru offers the opportunity to work on frontier AI systems that impact millions of users. Candidates should have at least three years of experience in ML infrastructure or distributed systems, with a strong background in GPU programming and training large models. The role is ideal for those passionate about system-level AI engineering and eager to contribute to India’s AI ecosystem at a high-impact organization.