Software Engineer, Infrastructure

Descript; Salem (OR); On-site; USD 150,000 - 230,000; Full time

Job summary

Descript is seeking a senior platform engineer to own the infrastructure that trains and serves models, runs GPUs, and powers AI tooling. You will shape reliability, cost, and security in production systems and collaborate across teams.

You’ll work with GCP, Kubernetes, Temporal, CI/CD, and infrastructure-as-code, defining on-call practices and improving observability, rollback mechanisms, and release processes while mentoring others.

Qualifications

• 8+ years building and operating production distributed systems.
• Experience with reliability, incident response, and post-incident reviews.
• Ability to define and use SLOs and error budgets to operate large systems.

Responsibilities

• Own our platform stack: GCP, Kubernetes, Temporal, GPU fleet, and deploy/rollback machinery.
• Own AI enablement substrate: training/inference pipelines and cost/reliability of serving models.
• Set infrastructure-as-code standards and build runbooks that explain system behavior.
• Improve observability, tests, and release practices to move fast without sacrificing quality.
• Provide architectural direction, code reviews, and mentoring to the team.

Skills

Distributed systems

On-call / incident response

Reliability engineering

SLOs & error budgets

Mentoring & architecture

Tools

GCP

Kubernetes

Temporal

Terraform

CI/CD

Job description

Platform owns the foundation the company runs on: compute and deployment, reliability and on-call, CI/CD and monorepo health, developer environments, the infrastructure that model training and inference run on, and the security boundaries around all of it. AI and agent tooling is the clearest example. You will contribute to how models are trained and served here, as well as how agents work inside our codebase: the environments they run in, the verification that makes their output trustworthy, and the review paths that keep it all legible. There is no industry standard and you’ll help form our opinions rather than inheriting one. Our users are other engineers. Expect to spend time with all the other engineering teams, understanding their needs and building a roadmap. The scope is large, so you'll be choosing what to leave alone as much as build. Architecture decisions here last: this is a small team covering a large surface. You'll have real room to decide things, and you'll stay close to the systems you decide about.

What you'll own

• Own our platform: GCP, Kubernetes, Temporal, the GPU fleet behind cloud export, and the deploy and rollback machinery everything ships through. You\'ll be in the on-call rotation, and we\'ll expect you to make it quieter and more actionable.
• The AI enablement substrate: GPU capacity, training and inference pipelines, and the reliability and cost of the systems serving models in production.
• Cost is an engineering constraint: infrastructure decisions have a number attached and you\'re expected to make smart trade-offs. As inference grows with usage, the metering and attribution behind those numbers sit with this team.
• Security comes with the systems you run : identity and access, secrets management, least-privilege boundaries, and supply-chain integrity.
• Make what you build legible. Infrastructure-as-code that explains itself. Runbooks and in-repo context written for someone jumping in to help. Observability that tells you what failed and why.
• Improve how the team learns and ships. Form hypotheses, instrument your work, release incrementally, and read results honestly. Strengthen the tooling, standards, tests, observability, and release practices that help the team move quickly without compromising quality.
• Raise the team’s technical ambition. Provide architectural direction, thoughtful reviews, mentoring, and clear human writing.

What you bring Required:

• You have 8+ years building and operating production distributed systems, or equivalent server-side engineering with a heavy infrastructure focus.
• You effectively leverage agents to multiply your impact and think critically about how and when to harness AI in your work.
• \'You\'ve run systems where failure was expensive, and your opinions about reliability and deployment come from consequences rather than reading.
• \'You\'ve carried a pager, commanded an incident, and rolled back before you understood why.
• You use SLOs and error budgets as operating tools.
• You\'ve used a major cloud provider and Kubernetes in production, with infrastructure-as-code as your default.
• You\'ve owned an architecture or migration, from planning through launch, whose consequences outlived the project, and you can say what you\'d do differently.
• You\'ve found important unowned work, scoped it, earned buy-in, and delivered it without being handed a spec.
• You build a minimal repro, read the logs, and write targeted checks to prove a fix works instead of trusting output.

Experience that helps:

• GPU and ML infrastructure: capacity planning, training or inference pipelines, serving cost and latency. If you haven\'t run a GPU fleet, experience with expensive capacity-constrained systems transfers well.
• Production security engineering: IAM, secrets, supply chain, least privilege. You don\'t need to have held a security title, but you should have owned these boundaries for systems you ran.