SlipstreamJobs tracks this role from the company's public career site. Apply directly on the employer's site.
Patronus AI is a frontier lab developing simulation research and infrastructure to accelerate progress toward human-aligned AGI. The company is behind influential AI evaluation research including FinanceBench, Lynx, SimpleSafetyTests, CopyrightCatcher, and Humanity's Last Exam. Founded by former researchers and engineers from Meta AI, Amazon AGI, and Google, the company serves foundation model labs and Fortune 500 enterprises like Adobe, backed by top-tier investors including Lightspeed Venture Partners and Stanford University.
As an AI Platform Engineer, you will sit between research and engineering, transforming research workflows into self-serve services for internal teams. The role is hands-on and delivery-oriented, focusing on systems and tooling that enable other teams' work rather than research itself.
Key responsibilities include:
- Building and operating internal platforms end-to-end (backends, storage, dashboards, CLIs, SDKs) with multi-tenancy, authentication, and access control
- Owning GPU cluster workload orchestration including submission, scheduling, provisioning, quotas, and placement
- Deploying and serving models on both managed inference providers and internally operated GPUs
- Building evaluation surfaces to maintain result comparability across time, teams, and models
- Establishing and hardening CI/CD pipelines, containerized workflows, and release safety
- Instrumenting platforms with logging, metrics, and alerting to catch service divergence
- Adding agent surfaces to the platform with auditable configuration
- Partnering with research engineers to productionize experiments and providing on-call support
Required qualifications include several years of hands-on production application development, strong backend engineering in production Python (APIs, async services, relational databases, object storage), hands-on experience deploying and serving models in production on both managed and self-operated GPU infrastructure with modern LLM serving stacks, experience running workloads on shared GPU clusters via schedulers like Slurm, and familiarity with cloud GPU infrastructure and CI/CD practices.
