Platform Engineer - AI Infrastructure

Ovii's Interpretation of the Role

Sarvam seeks a Platform Engineer to design and ship the control‑plane, scheduling, scaling and self‑service layers that power its sovereign AI GPU fleet. The role blends deep systems work in Kubernetes, Go/Python, and GPU‑specific constraints with a product mindset for internal ML engineers.

Role Snapshot

• Build AI platform control‑plane services
• Design scheduler & autoscaling for GPU workloads
• Implement multi‑tenant RBAC, quota and isolation
• Create CLI, SDK and APIs for ML engineers
• Develop observability, cost and networking tooling
• Provision clusters via IaC (Terraform / Crossplane)

Must-Have Requirements

• Go
• Python
• Kubernetes operators/controllers
• GPU platform constraints (MIG, gang scheduling, topology‑aware placement)
• building control‑plane services
• Kubernetes operator development
• GPU‑specific platform constraints

Nice-to-Have Signals

• Serving/inference platform experience
• GPU scheduling systems (Kueue, Volcano, Slurm)
• Deep Kubernetes networking (CNI, RDMA, SR‑IOV)
• On‑prem multi‑vendor GPU platform work
• Open‑source contributions to Kubernetes or GPU projects
• Terraform or Crossplane
• Observability stack (metrics, logging, tracing)
• serving/inference platform at scale
• multi‑tenant isolation mechanisms
• deep networking (CNI, RDMA)
• on‑prem GPU platform deployments

Work Setup

• Location: Bengaluru, India
• Work mode: ONSITE
• Employment type: Full-Time

Not Specified in JD

• Visa sponsorship
• Salary range
• Remote eligibility
• Education requirement
• Certifications
• Relocation
• Notice period
• Travel
• Security clearance
• Coding test
• Portfolio
• GitHub
• Writing sample
• Cover letter

What You'll Likely Work On

• Create the serving platform that turns model artifacts into scalable, multi‑tenant endpoints with rollout and traffic‑splitting capabilities
• Build autoscaling and elasticity layers for both training and inference, handling burst traffic, preemption and efficient GPU bin‑packing
• Develop the scheduler layer (Kueue, Volcano, Slurm or custom controllers) with gang scheduling, priority, quota enforcement and topology‑aware placement
• Implement multi‑tenant isolation, RBAC, quota and self‑service tiers for MIG/MPS and time‑slicing
• Engineer platform‑level networking, including CNI configuration, RDMA/SR‑IOV exposure and multi‑cluster connectivity
• Provide fleet‑scale observability and cost tooling for metrics, logs and traces
• Design storage abstractions over parallel filesystems, caching tiers and data‑locality‑aware volume provisioning
• Deliver developer experience via CLI, SDK and APIs, plus documentation for self‑service job submission

Good Fit If You Have

• Prior experience building a serving or inference platform at scale
• Hands‑on work with GPU scheduling systems such as Kueue, Volcano or Slurm
• Deep knowledge of Kubernetes networking internals and RDMA/SR‑IOV
• Experience with on‑prem multi‑vendor GPU environments