Infrastructure Engineer
Full-time · Remote
[About Modern Relay]
Modern Relay builds the infrastructure companies use to build the context layer their AI agents work from. Our engine, Omnigraph, keeps a company's context as one shared, versioned graph in its own storage. It's written in Rust and open source.
We’re backed by Point Nine, Emerge and Amino Collective, and by angels including Charlie Songhurst (Meta board member), Martin Gould (Spotify), Thomas Clozel (Owkin) and Sarah Drinkwater (Google, Common Magic).
[The role]
Omnigraph Cloud, now in private preview, gives each customer a dedicated Omnigraph cluster with its own compute and object storage, and runs it for them: upgrades, schema migrations, backups, compaction and monitoring. A control plane handles SSO, scoped tokens, Cedar policy and audit.
You'll build and run the infrastructure behind it, and the same deployment in customers' own cloud accounts. As the number of customers and clusters grows, you'll keep improving it: making it more reliable, faster and cheaper to run, and taking manual work out of operating it.
[What you'll work on]
- Provision dedicated clusters and their storage on AWS, Azure and GCP, in our accounts and in customers' own
- Roll out releases and schema migrations across clusters without interrupting customers
- Run backups, compaction and recovery for graphs stored as Lance files in object storage
- Set up monitoring and alerting for every cluster, and take part in on-call
- Keep customers isolated from each other, and build the control plane's identity, tokens and audit trail
- Scale the fleet as customers grow: capacity planning, autoscaling, and automating every step that is still done by hand
- Measure where time, failures and cost come from, and fix the biggest causes first
[You might be a fit if]
- You've run production infrastructure through incidents, from report to root cause, fix and follow-up
- You know at least one of AWS, Azure or GCP well, including its networking, identity and object storage
- You've managed infrastructure as code and automated deployments and upgrades
- You can reason about what happens to stored state when a machine, a network or a deploy fails, and how the system recovers
- You've scaled a system as its load or number of customers grew, and can show what you measured and what you changed
- You build mainly with coding agents, and accept their output only on tests, measurements and review
[Nice to have]
- Terraform, Pulumi or Kubernetes in production
- Capacity planning, autoscaling or cloud cost work at scale
- Running a service where each customer gets dedicated infrastructure
- Identity and access work: SSO, SCIM or policy engines such as Cedar