SlipstreamJobs tracks this role from the company's public career site. Apply directly on the employer's site.
OpenAI Consumer Devices is building next-generation products that bring powerful AI into everyday life. The team combines researchers, engineers, designers, and operators working on zero-to-one challenges at the intersection of hardware, software, and AI.
As an Operating Systems Engineer focused on on-device inference, you will design, develop, and ship the OS stack that makes advanced AI capabilities reliable, responsive, and energy efficient on consumer devices. Your work spans OS services and frameworks, inference runtime integration, model fitting, scheduling, and performance and power management. You'll partner with research to adapt models to device constraints, make design decisions across the stack, and carry solutions from early exploration through integration and production.
Key responsibilities include:
- Building the inference platform: Design and implement maintainable OS services, frameworks, and clear interfaces for inference execution, model loading and lifecycle, and resource management.
- Fitting models to device constraints: Partner with researchers on quantization, runtime integration, and memory optimization to meet memory, compute, and energy budgets while evaluating model quality and product behavior.
- Coordinating system resources: Develop scheduling and resource policies that balance inference with other device activity, preserving responsiveness within latency, memory, battery, and thermal constraints.
- Advancing performance and power management: Develop and validate execution strategies that adapt to workload needs, available resources, and changing device conditions.
- Debugging across the stack: Use tracing, profiling, and structured debugging to investigate correctness, concurrency, performance, and reliability issues across models, inference runtimes, and OS components.
- Measuring and validating improvements: Build diagnostic tools, instrumentation, representative device workloads, and automated tests to demonstrate repeatable performance and energy gains on physical devices, catch regressions, and validate sustained use.
- Bringing capabilities to production: Collaborate with research, hardware, firmware, platform, and product engineering teams to turn emerging model capabilities into maintainable systems and reliable shipped features.
REQUIREMENTS:
Minimum qualifications:
- Substantial hands-on experience designing, developing, and debugging operating system components, system services, or performance-critical platform software.
- Proficiency in C++ for systems development, including concurrent programming, memory ownership, and resource lifetime management.
- Hands-on experience integrating or optimizing inference runtimes or machine learning workloads in resource-constrained environments.
- Strong understanding of scheduling, memory management, and how workloads compete for shared system resources.
- Experience diagnosing complex system behavior and delivering performance or power improvements supported by repeatable measurements.
- Ability to work across disciplines, translate research and product needs into system requirements, and explain technical decisions and tradeoffs clearly.
Preferred qualifications:
- Delivered end-to-end on-device inference in shipped products, from model adaptation and runtime integration through OS support and production debugging.
- Adapted models for deployment through quantization, compression, or related techniques, balancing quality, execution cost, and memory requirements.
- Optimized inference across CPUs, GPUs, or neural accelerators, accounting for data movement, synchronization, and execution placement.
- Proficiency in Rust for systems programming.
- Track record of taking ownership of complete solutions and following problems across model, runtime, and operating system boundaries.
- Ability to challenge assumptions, develop new approaches when existing techniques fall short, and use experiments and measurements to guide decisions.
- Experience working constructively across disciplines and making careful tradeoffs among model quality, performance, power, reliability, security, and maintainability.
