Member of Technical Staff (Site Reliability Engineer, Cloud Infrastructure)

Perplexity; California; On-site; USD 180,000 - 240,000; Full time

Job summary

Perplexity is seeking a Site Reliability Engineer to own compute, networking, and storage at scale. You will ensure availability, latency, and cost across consumer, computer, API, and enterprise products while the platform evolves rapidly.

You will run our Kubernetes fleet, design resilient VPCs, manage data stores, and drive infrastructure as code with Terraform and CI/CD. You will lead incident response, define SLOs, and run game days to prove recovery.

Qualifications

• Deep AWS expertise including VPC, IAM, EC2, EKS, ELB, Route 53, S3, EBS and cost controls.
• Production Kubernetes at scale with cluster upgrades, CNI, CoreDNS, autoscalers like Karpenter, and debugging at 3 a.m.
• Terraform mastery across many accounts, managing state and large refactors without production impact.
• Strong networking fundamentals (TCP/IP, DNS, TLS, load balancing, routing).
• You write real code in Python, Go, or Rust for automation, tooling, and controllers.
• Observability with Datadog, Prometheus, or Grafana; turning metrics/logs/traces into meaningful SLOs/alerts.
• Track record of leading production incidents and durable root-cause fixes.
• You own problems end-to-end and thrive when the path forward isn't laid out.

Responsibilities

• Own production at scale: manage compute, networking, and storage foundation behind Perplexity's products and ensure availability, latency, and cost.
• Run our Kubernetes fleet: operate and upgrade EKS clusters, tune scheduling, autoscaling, DNS, and service networking under bursty loads.
• Engineer the network: design VPCs, Transit Gateway, PrivateLink, load balancing, DNS and cross-account connectivity.
• Keep data durable and fast: oversee storage (EBS, S3, DynamoDB) and databases/backups with tested recovery.
• Manage everything as code: IaC with Terraform and CI/CD; create reusable modules and safe rollback guardrails.
• Prepare for failures: define SLOs, build actionable alerts, design for regional failures, and run game days.
• Lead incident response: drive incidents to resolution, write blameless postmortems, and implement durable fixes.
• Scale ahead of demand: capacity planning, quotas, reservations, and cost optimization for launches.
• Automate away toil: build self-serve tooling to reduce manual provisioning and remediation.

Job description

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.

Responsibilities

• 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.

Qualifications

• 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.

Additional experience we value

• 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.