Senior DevOps Engineer

AWS Infrastructure, Kubernetes & Production Operations

Mumbai (On-site) | Full-time | 5-8 years

About the Role:

Unico Connect is an AI-first technology partner that builds custom mobile, web, and AI products for clients across multiple geographies.

We are hiring a Senior DevOps Engineer for a dedicated client engagement building an AI platform that turns a written prompt into a working, deployed web application.

The platform runs entirely on our own Kubernetes on AWS in Mumbai: multiple clusters, self-hosted Kafka, PostgreSQL, Redis and a complete observability stack, with untrusted AI-generated code executing inside isolated sandboxes.

This role owns that infrastructure end to end.

The mandatory requirement is hands-on production ownership of self-managed Kubernetes and self-hosted stateful services.

This is not a managed-services role: almost nothing here is a click in a console, and there is no platform team above you to escalate to.

You will pair daily with the Tech Architect who owns the architecture, and you own its execution, which covers clusters, networking, pipelines, observability, isolation, and the monthly cloud bill as a number you can explain line by line.

The role carries a deliberate path into LLM infrastructure over the first year, and we are saying so up front because we want someone who wants that direction.

A typical week includes rehearsing a cluster upgrade in a lower workplace, tracing an unexplained cost line to its source, tuning autoscaling so non-production shuts down cleanly overnight, and hardening the boundary that keeps AI-generated code away from everything else.

Responsibilities:

Kubernetes Platform Ownership

Own multiple EKS clusters separated by trust boundary, covering platform services, messaging, sandboxed build environments, and hosting for customers' published applications.

Node pools, taints and tolerations, resource requests versus limits, pod disruption budgets, and the scheduling decisions that determine whether a cluster is efficient or merely large.

Plan and execute version upgrades without customer-visible downtime.

Infrastructure as Code

Everything in Terraform, with no console changes that are not in code.

Build reusable cluster modules rather than copies.

Keep state and reality in sync, and treat drift as a defect.

CI/CD and Release Engineering

Own the GitHub Actions pipelines and self-hosted runners: OIDC-based AWS authentication with no stored credentials, image builds, and rollouts across development, QA and UAT.

Move manifest templating from scripted substitution to Kustomize or Helm.

Make a rollback as routine as a deploy.

Network Architecture and Workload Isolation

VPC and subnet design, ingress through load balancers and service mesh, and eBPF-based network policy.

The platform runs code we did not write, so the isolation boundary is a primary deliverable: default-deny egress, kernel-level sandboxing for untrusted workloads, no platform credentials inside tenant pods,



and one project per pod with nothing shared between customers.

Observability

Operate a self-hosted stack of Prometheus, Grafana, Loki, Tempo, OpenTelemetry and Alertmanager.

Build alerts engineers trust rather than mute.

Know what belongs in a metric, a log and a trace, and keep metric cardinality under control as tenant count grows.

Autoscaling and Capacity

Node autoscaling with Karpenter, and workload autoscaling on the signal that actually matters, which is queue depth or a schedule rather than CPU wherever CPU misleads.

Implement scale-to-zero across non-production and warm-pool patterns where cold starts are unacceptable.

Size capacity from measurement, not assumption.

Cost Operations

Own the cloud bill as an engineering artifact.

Attribute spend per cluster and per environment, right-size against real utilisation, apply Spot where interruption is safe and never where it is not, and investigate any line you cannot explain.

Instance family selection is a cost decision: match the family to the resource that binds first.

Self-Hosted Stateful Services

Operate Kafka, PostgreSQL and Redis in-cluster, covering persistence, replication, failover, upgrades, and backups you have actually restored from.

Rehearse the restore and record how long it took.

Security Posture

IAM and workload identity, migration away from static credentials, secret management, web application firewall rules, and closing the gaps you find.

Credential rotation and least privilege are ongoing work, not a one-time project.

AI Infrastructure (Growth Path)

Within the first year, take on the infrastructure serving AI workloads: multi-provider LLM gateway and routing, token and cost attribution, rate-limit-aware capacity planning, LLM tracing with Langfuse, vector databases such as pgvector and Qdrant, and self-hosted embedding models including GPU nodes on Kubernetes.

This is inference operations, not model training.

AI-Assisted Engineering Discipline

Use Claude, Cursor, and similar tools day to day for infrastructure code, scripts, and pipelines.

Set the team standard for safe use, review, and validation of AI-generated infrastructure before it ships.

Requirements:

Hands-on Production Ownership of Self-Managed Kubernetes (Mandatory)

Must have personally operated a production Kubernetes estate with operational responsibility including oncall, incident response, and upgrade ownership.

You understand why a cluster ends up over-provisioned and how to fix it.

Managed-platform-only experience, or Kubernetes limited to deploying into a cluster someone else runs, does not qualify.

5+ Years of Overall Engineering Experience





With at least 3 years in DevOps, SRE, or platform engineering.

This is not an ML science role.

A strong infrastructure background with genuine appetite for AI infrastructure is exactly the profile we want.

Proven Ability to Learn an Unfamiliar Technology to Production Depth

We are hiring for trajectory as much as for current tooling, so this carries real weight.

Be ready to walk us through something you picked up from scratch and then owned in production: what you got wrong in the first attempt, how you found out, and what you would do differently.

A candidate with a shorter tool list and a strong record of learning will be preferred over a longer list learned only at the surface.

Self-Hosted Stateful Services in Production

One or more of Kafka, PostgreSQL, Redis, or Elasticsearch operated by you, not by a managed service.

Backups, restores, failover, and version upgrades.

This matters more than any other line here, because almost everything we run, we run ourselves.

Infrastructure as Code Depth

Terraform in anger, including modules, state management, and recovering from drift.

Docker and image optimisation.

AWS at Depth

EKS, EC2, VPC, IAM, S3, RDS or Aurora, and load balancing.

You can read a bill and explain where the money went.

Networking Depth

VPC and subnet design, NAT and egress paths, load balancer behaviour, DNS, TLS, and Kubernetes networking.

Able to reason about address exhaustion and routing limits before they are hit rather than after.

Observability Depth

Production experience with Prometheus, Grafana, and a log or trace backend.

Comfortable instrumenting with OpenTelemetry.

Able to define alerts that are actionable and to resist the cardinality that makes a metrics system collapse.

CI/CD and Deployment Automation

Strong with GitHub Actions or GitLab CI, including self-hosted runners and short-lived cloud credentials.

Experience making deploys and rollbacks safe and routine.

Cost Management for Cloud Infrastructure

Demonstrated ownership of a cloud bill: right-sizing, reserved capacity or savings plans, Spot strategy, and tracing an anomaly to its cause.

Reliability Engineering

Comfortable defining SLOs and being accountable for them.

Experience with graceful shutdown, connection draining, disruption budgets, and long-lived connections such as WebSockets that autoscaling can otherwise sever.

Nice to Have:

- arm64 and Graviton estates
- eBPF networking with Cilium
- container runtime isolation such as gVisor, Kata, or Firecracker
- Kafka operations including partition and consumer-group management
- service mesh with Istio
- GPU workloads on Kubernetes
- hands-on familiarity with LLM observability tooling such as Langfuse or LangSmith
- India data-protection and residency requirements under the DPDP Act
- AWS certifications

Skills:- Kubernetes, Amazon Web Services (AWS), EKS, Terraform, Docker, Kafka, PostgreSQL, Redis, GitHub Actions, helm, prometheus and grafana

📌 Senior DevOps Engineer (Mumbai)
🏢 Unico Connect Private
📍 Mumbai