Magic

Build the software, automation, and infrastructure that defend Magic’s frontier AI systems and engineering organization from traditional and emerging AI cyber threats. This hands-on role combines security engineering, infrastructure protection, secure sandboxing, threat response, and LLM-native attack defense.

Build and operate distributed systems that train long-context models across massive GPU clusters. You’ll improve training performance, reliability, fault tolerance, and reproducibility at frontier scale.

Build and operate the distributed inference and RL systems that make Magic’s long-context models fast, reliable, and scalable. You’ll optimize serving performance, post-training workflows, and production infrastructure across GPU, networking, and storage layers.

Lead the developer-experience and data-tooling infrastructure that helps Magic’s pre-training data team work faster. Build internal tools, dashboards, CLIs, web UIs, and workflow improvements with broad ownership and rapid feedback loops.

Build end-to-end web products and developer tools on top of Magic’s long-context models, spanning frontend, backend, APIs, and model integration. You’ll turn advanced model capabilities into intuitive, reliable user experiences while owning product work from concept through iteration.

Build the datasets, evaluation frameworks, reward signals, and environments that drive reinforcement-learning post-training for long-context models. You’ll connect software, data, infrastructure, and research experiments to measurable capability improvements.

Join Magic in a unique role focused on building safe AGI through innovative research and code generation.

Build and operate the large-scale GPU platform that powers Magic’s model training and inference workloads. You’ll own Terraform-based infrastructure, Kubernetes clusters, and the reliability of distributed compute, networking, and storage systems.

Design and optimize high-performance GPU and accelerator kernels powering long-context AI training and inference. You’ll own kernel development through production while tackling challenging memory, data movement, and throughput problems for frontier AGI systems.