Note: The job is a remote job and is open to candidates in USA. OpenRouter is an AI routing and infrastructure layer that helps enterprises access, manage, and optimize large language models across providers. The Site Reliability Engineer will own the operational health, observability, reliability, and quality of OpenRouter’s provider supply, including incident response, failover, automation, capacity planning, and launch readiness.
Responsibilities
- Provider health and observability. Build and own monitoring for every provider and endpoint: latency, throughput, error rates, uptime, and output correctness. Set SLOs per provider tier and alert on them
- Detection and failover. Improve how quickly we detect degraded endpoints, and work with the routing team so traffic shifts away from them automatically
- Incident response. Own on-call for provider incidents: triage, mitigate, communicate with providers, run postmortems, and drive follow-ups to closure
- Provider accountability. Turn telemetry into scorecards and SLO reporting that providers act on, and be the technical escalation point when a provider's endpoint is misbehaving
- Quality regression detection. Build continuous canaries and evals that catch silent regressions (quantization changes, broken tool calling, truncated streams, pricing or usage-reporting mismatches), not just outright downtime
- Automate the toil. Replace manual provider-ops work (disabling endpoints, capacity changes, deprecations, rate-limit tuning) with safe, auditable tooling
- Capacity and launch readiness. Build tooling to load-test endpoints before big launches so day-zero traffic doesn't take them down
Skills
- 4+ years in SRE, production engineering, or infrastructure roles running high-traffic, customer-facing systems
- Strong with observability tooling and practice: metrics, tracing, logs, SLOs/error budgets, alerting that is always actionable
- Capable software engineer who prefers writing tools over executing runbooks. TypeScript and/or Python
- Experienced with distributed systems failure modes: timeouts, retries, backpressure, partial outages, noisy neighbors
- Calm, clear incident commander who communicates well with external partners under pressure
- Understands, or is eager to learn deeply, how LLM inference is served: streaming, tool calling, prompt caching, throughput/latency tradeoffs, and how provider APIs differ
- Experience at an inference provider, model lab, GPU cloud, or API gateway/CDN company
- Experience with our stack: TypeScript, Cloudflare Workers, Postgres, ClickHouse, GCP, Vercel
- Background in routing, load balancing, or traffic management systems
- Experience with evals or synthetic monitoring for ML systems
Benefits
- Remote work arrangement
Company Overview
Company H1B Sponsorship