Member of Technical Staff - Reliability Engineering

ABOUT US: Fireworks is the platform for specialized intelligence, enabling companies to build, train, and serve AI models tailored to their own data, workflows, and products. Founded by the team behind PyTorch and backed by AMD, Atreides, Benchmark Capital, Index Ventures, Lightspeed, NVIDIA, Sequoia Capital, and TCV, Fireworks powers production AI with hundreds of state-of-the-art open models across text, image, embedding, audio, and multimodal workloads. Today, Fireworks is a Series D company valued at $17.5 billion, bringing together an ambitious, collaborative team that's building the future of enterprise AI. About the Role Fireworks AI is one of the industry leaders in inference and training for open models. Open models are how the rest of the world gets to build on frontier AI without handing the keys to a single vendor, and our job is to make them fast, cheap, and dependable enough that this is a real choice. That work is systems work: GPU scheduling, kernel and runtime performance, networking, storage, Linux. We serve over 40 trillion tokens a day doing it. Reliability Engineering makes sure that platform runs dependably as it grows. You will work across cloud infrastructure, AI systems, and product teams to make sure the pieces fit together, fail gracefully, and hold up under load. How We Think About Ownership - You own the bar. You define what "reliable" means at Fireworks: SLOs, error budgets, production readiness, on-call expectations. Then you drive adoption across engineering. - You own the process and the tooling. Incident management, postmortems, observability standards, failure testing, guardrails, and automation are yours end to end. - Every team owns the reliability of what they build. You make that ownership practical. Structured logging and aggregation, metrics and tracing that work the same way everywhere, alerting that routes to the right owner, dashboards that answer "why is this slow." - You choose where the leverage is. You have a wide view of the platform and the latitude to spend your time where it changes outcomes most. Responsibilities - Define reliability standards: SLOs, error budgets, production readiness criteria. Not written in a vacuum: you instrument the systems and read the real telemetry the numbers come from. - Own the reliability toolchain: Logging and telemetry pipelines, alerting standards, failure injection, load testing, self-healing automation, and AI-assisted investigation tooling. - Keep customer experience from falling through the cracks: Per-service reliability is necessary but not sufficient. A customer can hit a bad experience while every system sits inside its SLO. You make sure those failures get an owner and a fix. - Own the seams: The hardest failures live between systems: retries that amplify load, timeouts that do not compose, dependencies nobody mapped. You find them before customers do and drive fixes through the teams that own them. - Run incident management: Coordinate live production issues, run blameless postmortems, and track follow-ups to completion. - Reduce toil: Automate repetitive operational work so growth does not turn into an unsustainable on-call load. - Partner across the org: Cloud infrastructure on capacity and multi-region risk, inference and training on failure modes in the serving and training stacks, performance on zero-downtime rollouts, product and control plane on customer-facing reliability. Qualifications - Systems fundamentals: 5+ years with Linux internals, system performance troubleshooting, and networking fundamentals (TCP/IP, HTTP, gRPC). - Software engineering: 5+ years in Python, Go, C++, or Rust, writing production-grade tools and systems code. - Cloud-native operations: Operating and debugging Kubernetes, Terraform, and Docker in high-throughput production. - Distributed systems: High-throughput control planes, microservices, or multi-region setups. - Reliability fundamentals: Fault-tolerant design, SLO/SLA management, automated failover, high-availability architecture. - Influence without authority: You can get other teams to adopt a standard through credibility and useful tooling rather than mandate. - Breadth over comfort: Willingness to dig into unfamiliar parts of the stack when a problem crosses boundaries. - Education: Bachelor's or Master's in Computer Science, Computer Engineering, or equivalent practical experience. Preferred Qualifications - Observability tooling: Prometheus, Grafana, OpenTelemetry, and alerting people actually act on. - GPU and ML infrastructure exposure: GPUs, inference serving, or distributed training. - AI-assisted operations: Building agents or LLM-based tooling for investigation, triage, or automation. - Open source background: Contributions to infrastructure, systems, or ML serving projects. - Startup agility: Comfortable where pragmatism and teamwork matter more than process. WHY FIREWORKS? - Solve Hard Problems: Tackle challenges at the forefront of AI infrastructure, from low…