Want to know if this job is worth applying to?
San Francisco, California, United States
Onsite
Perplexity serves hundreds of millions of queries a month. Every answer, every Perplexity Computer task, and every enterprise deployment runs on production infrastructure owned by the Cloud Infrastructure team: a fleet of Kubernetes clusters across a multi-account AWS estate, private networks that reach hyperscalers and the neoclouds serving our models, and the storage underneath our user data and agentic workloads. We're looking for SREs who treat reliability as a software problem. You'll own compute, networking, and storage at scale and keep production stable while everything built on top of it is changing fast.
Own production at scale. Run the compute, networking, and storage foundation behind Perplexity's consumer, Computer, API, and enterprise products, and own its availability, latency, and cost.
Run our Kubernetes fleet. Operate and upgrade the EKS clusters that host web services, search, datastores, and secure sandboxes. Tune scheduling, autoscaling, node lifecycle, DNS, and service networking so clusters stay healthy under bursty, fast-growing load.
Engineer the network. Design and operate VPC architectures, Transit Gateway and PrivateLink connectivity, load balancing, DNS, and egress across accounts, regions, and cloud providers.
Keep data durable and fast. Own the storage layer (EBS, S3, DynamoDB, and the managed databases and caches our services depend on), including performance, durability, backups, and restores that have actually been tested.
Manage everything as code. Run infrastructure through Terraform and CI/CD. Build reusable modules and guardrails, and make every change reviewable, repeatable, and safe to roll back.
Prepare for failures. Define SLOs, build alerts that someone can act on, design for zone and regional failure, and run game days and disaster-recovery exercises that prove recovery works.
Lead incident response. Be on the critical path when things break. Drive incidents to resolution, write blameless postmortems, and land the fixes so the same failure never pages anyone twice.
Scale ahead of demand. Own capacity planning, quotas, reservations, and cost efficiency so product launches and traffic spikes never wait on infrastructure.
Automate away the toil. Build tooling, increasingly agent-driven, that turns provisioning, upgrades, and remediation into self-serve, hands-off workflows for every engineering team.
We expect you to have real depth in most of these:
Deep AWS expertise. You know VPC, IAM, EC2, EKS, ELB, Route 53, S3, EBS, etc. well past the console, including their limits, failure modes, and the bill that comes with them.
Production Kubernetes at scale: cluster upgrades, CNI, CoreDNS, autoscalers like Karpenter, and debugging a misbehaving node at 3 a.m. Writing Helm charts alone doesn't count.
Terraform mastery. You've designed module structures, managed state across many accounts, and run large refactors and imports without taking production down.
Strong networking fundamentals (TCP/IP, DNS, TLS, load balancing, routing).
You write real code in Python, Go, or Rust for automation, tooling, and controllers, and you'd rather automate a runbook than follow it.
Fluency with observability tools like Datadog, Prometheus, or Grafana, and the judgment to turn metrics, logs, and traces into SLOs and alerts that matter.
A track record of leading production incidents and turning root-cause analysis into durable fixes.
You own problems end-to-end and do well when the path forward isn't laid out for you.
Multi-region active-active or failover architectures, and disaster-recovery programs with real recovery targets.
Proxies, service meshes, and kernel-level networking: Envoy, Istio, Cilium, or eBPF.
GitOps and progressive delivery: Argo CD, Flux, or similar.
Enterprise and compliance environments: single-tenant or BYOC deployments, KMS/BYOK, SOC 2 or similar controls.
Running stateful systems in production: Postgres, Redis, or search engines like Vespa or Elasticsearch.
Multi-cloud or neocloud operations (GCP, CoreWeave, or similar).
FinOps: cost attribution, Savings Plans and reservations, and making infrastructure cheaper without making it fragile.
