Want to know if this job is worth applying to?
San Francisco, California, United States
Onsite
Vizcom is an AI design tool for industrial and product designers. Teams at GM, Ford, Nike, adidas, Hasbro and EA use it to go from sketch to render to 3D in minutes. We're Series B, 300+ enterprise customers, small team, shipping fast.
Under the hood we're an inference orchestration platform with a CRUD app attached. Most of what breaks lives in how we queue, retry, route and recover work running on external GPU providers (Fal, Vertex, Modal, ByteDance). Our biggest customers are subscribed to our status page, so reliability is a sales lever, not just an eng goal.
You're the second SRE. Sam joined in September and has been rebuilding our observability in Datadog and Cloudflare. You'll split the platform with him and help turn SRE into a real team.
The job is to let product engineering ship every day without breaking things. It's open-ended and high-impact: find the problem, fix it, ship it, make sure it stays fixed. If you like being the person who actually knows how the whole system works, you'll like this. You report to Kaelan (CTO).
SLOs that mean something. Define "reliable" per surface (API, websockets, jobs, renders), measure it, and hold us to it. We have the dashboards started. They need teeth.
Inference reliability. Provider failover, model fallbacks, queue priority for enterprise traffic, and degrading gracefully when a GPU vendor blips. The multi-provider router exists. Make it predictable under partial failure.
Kubernetes and infra. Probes, rollouts and rollbacks, graceful shutdown, autoscaling headroom, resource limits under bursty inference load. Everything is Pulumi. Keep it that way.
Queues and jobs. At-least-once is the reality. Idempotency, poison-pill containment and stalled-job recovery should be defaults, not patches.
Ship-safe pipeline. Canaries, feature-flag rollouts, lock-aware migrations. A bad deploy should be boring to roll back.
Incidents. Lead Sev1/Sev2 response, including status page and enterprise comms. Run blameless postmortems and make sure the same thing doesn't break twice.
Security and compliance infra. Work through our pentest findings (AWS account isolation, secrets management) and keep SOC 2 / ISO evidence from being a fire drill.
Cost. Reliability and COGS are the same lever here. Help us spend GPU money on purpose.
Learn the system and get onto on-call.
Own one pentest remediation end to end (likely secrets management).
Ship SLOs plus alerting for the top 3 customer-facing surfaces, with pages that are actually worth waking up for.
Run one postmortem to durable fixes shipped within 30 days.
Node/Koa, GraphQL (PostGraphile), Postgres, Redis, BullMQ, Kubernetes on AWS, Pulumi, GitHub Actions, Cloudflare, Datadog, Statsig, WorkOS. Inference runs on Fal, Vertex, Modal and ByteDance. One major enterprise customer runs on a fully isolated stack.
We also run an in-house SRE agent that posts daily health digests and flags regressions. You'll help decide what it should and shouldn't do.
5+ years running production systems, ideally at a startup that was growing fast. Flat load on a system that was already hardened doesn't count for much here.
Strong on observability, incident response and root cause analysis. You've been the incident lead on a real Sev1.
Comfortable with Kubernetes, networking and AWS.
Real infrastructure-as-code experience (Pulumi, Terraform or similar).
You can read and fix application code (we're TypeScript), not just the infra under it.
You think in failure modes and blast radius by default.
You own problems end to end. "I filed a ticket" is not the end.
Clear written communication. Postmortems and RFCs are part of the job.
Has run GPU or ML inference workloads in production.
Has worked with queue-heavy systems (BullMQ, Sidekiq, Celery, SQS).
Has done SOC 2 or ISO evidence work, or pentest remediation.
Has set up canary deploys and feature-flag rollouts.
Level: Senior SRE.
Location: San Francisco preferred.
Comp: TBD base plus equity.
On-call: You carry a pager, on a rotation with Sam and senior eng. You'll help shape the rotation.
Visa: Not sponsoring for this role.
Recruiter screen (30 min)
Hiring manager chat with Kaelan (30 min)
Coding: backend/SRE job-queue problem (60 min)
Incident deep-dive: walk us through one you led (45 min)
Onsite or virtual panel, including a systems design session on inference failover
References and offer