Infrastructure SRE - HPC
About this job
Sarvam AI is creating India's complete sovereign AI platform by focusing on research, models, infrastructure, and applications to make artificial intelligence truly useful for the country. As an Infrastructure SRE focusing on High-Performance Computing, you will manage a large, multi-vendor GPU fleet handling demanding concurrent workloads like multi-week training jobs and low-latency inference services. A typical day involves ensuring high fleet reliability, troubleshooting complex distributed system failures across parallel filesystems, RDMA fabrics, and networks, building internal operational tooling, and collaborating closely with platform and machine learning teams to keep large training runs active and maintain predictable serving latencies.
Responsibilities
- Manage the end-to-end GPU fleet across provisioning, capacity, observability, and health for training and serving.
- Participate in the on-call rotation, author reliable runbooks, and conduct postmortems to implement permanent fixes.
- Develop bespoke internal tooling instead of relying solely on standard off-the-shelf systems.
- Collaborate with ML and platform engineering teams to maintain the uptime of large training runs and ensure stable serving latency.
- Triage and route complex cross-domain incidents spanning storage, networking, and cluster hardware.
- Focus on one of five specialized areas including high-performance storage, RDMA networking, GPU reliability, Kubernetes platform reliability, or workload reliability.
Requirements
- 5+ years of experience in site reliability or infrastructure engineering.
- 2+ years of hands-on experience managing GPU clusters at scale.
- Proven ownership of critical infrastructure during on-call duties, backed by actionable postmortems.
- Strong programming skills in Python or Go for building and maintaining internal tools.
- Working fluency across storage, networking, GPU systems, Kubernetes, and workloads to triage and route issues effectively.
- Deep domain expertise in at least one of the designated areas of focus.
Good to have
- Experience managing Slurm and Kubernetes hybrid environments.
- Background in on-premise GPU deployments, including data center coordination for power, cooling, and cabling.
- Familiarity with Indian NCPs, DGX SuperPOD, Lambda, CoreWeave, or NeevCloud.
- Experience implementing multi-tenant GPU isolation techniques like MIG, MPS, or time-slicing in production.
Who should apply
This role suits experienced infrastructure and site reliability engineers who possess deep specialized knowledge in one core area such as distributed storage, RDMA networking, GPU reliability, Kubernetes, or workload reliability, alongside a strong baseline fluency across the others. Candidates must have direct experience running GPU clusters at scale, strong coding skills in Python or Go, and a track record of handling high-pressure on-call duties. Freshers should not apply as this position requires professional engineering experience.
Frequently asked questions
What is the primary mission of the Infrastructure SRE - HPC role at Sarvam AI?
The role focuses on maintaining the health, reliability, and scale of Sarvam AI's multi-vendor GPU fleet, supporting heavy training jobs and low-latency inference services.
What core technical background is required to join Sarvam AI as an Infrastructure SRE - HPC?
Candidates need 5+ years in infrastructure or SRE, including 2+ years operating GPU clusters at scale, proficiency in Python or Go, and deep depth in one of five technical focus areas.
Where is the Infrastructure SRE - HPC position located?
The position is based in Bengaluru.
Can freshers apply for the Infrastructure SRE - HPC opening at Sarvam AI?
No, freshers cannot apply because the role requires at least 5 years of professional infrastructure or SRE experience with specific GPU cluster management background.
Rojgar Point lists this job for information. Always apply on the official company page and never pay anyone for a job.