Principal AI SoC Runtime Software Architect

Define the SoC-wide execution model for coordinating workloads across heterogeneous compute and media engines, including dependency management, resource arbitration, priority and QoS, and concurrent pipeline behavior.

Set the architecture and technical direction for the AI inference runtime, including compiled-model execution, integration with industry-standard AI execution frameworks, and interfaces to applications and the broader Velaura SDK.

Own the end-to-end dataflow architecture for sensor-to-application pipelines, ensuring that camera, media, preprocessing, inference, and postprocessing components operate as an efficient and coherent system.

Define the SoC-wide memory and buffer-sharing architecture across user space, the kernel, and heterogeneous hardware engines, establishing clear ownership, coherency, isolation, synchronization, and lifecycle semantics.

Define the division of responsibility and interface contracts among the runtime, kernel drivers, firmware, and hardware engines, including execution, completion, telemetry, fault management, and recovery semantics.

Partner with the compiler team to define the compiler–runtime contract, ensuring compiled artifacts contain the metadata and execution information needed for the runtime to load, validate, execute, profile, and maintain compatibility across releases.

Establish a system-wide observability and performance architecture that correlates behavior across software and hardware layers and enables optimization against latency, throughput, bandwidth, power, utilization, and predictability goals.

Define the runtime resilience and validation architecture, including fault-containment and recovery policies, architecture-level acceptance criteria, and qualification across correctness, concurrency, compatibility, performance, and sustained workloads.

Extensive experience designing and building production runtime systems, embedded middleware, multimedia frameworks, or other performance-critical systems software.

Strong C/C++ programming skills and demonstrated ability to architect and contribute hands-on to production runtime software spanning application-facing APIs, user-space libraries, and low-level driver, firmware, and hardware interfaces.

Strong understanding of heterogeneous and asynchronous execution, including command submission, queues, events, dependencies, synchronization, concurrency, scheduling, and resource management.

Strong understanding of device memory, DMA, IOMMU/SMMU, cache coherency, memory mapping, shared buffers, buffer lifetimes, and kernel/user-space memory interfaces.

Experience optimizing end-to-end data movement and execution across multiple hardware engines rather than focusing solely on individual kernels or accelerator performance.

Experience designing stable runtime APIs with well-defined compatibility, versioning, error handling, diagnostics, and recovery behavior.

Demonstrated ability to debug complex cross-layer correctness and performance problems using disciplined, data-driven methods, profiling, and tracing.

Demonstrated technical leadership across component and organizational boundaries, including translating system requirements into clear architectures, interfaces, implementation guidance, and validation strategies.

Direct experience developing or extending AI inference runtimes or execution providers, such as ONNX Runtime, TensorRT-like runtimes, OpenVINO, TensorFlow Lite delegates, Qualcomm QNN/SNPE, TVM runtimes, or comparable systems for NPUs, GPUs, DSPs, or other accelerators.

Linux kernel development or upstream contribution experience involving device, accelerator, media, or shared-memory subsystems.

Experience with robotics, autonomous systems, edge AI, ROS 2, camera pipelines, ISP integration, V4L2/media, GStreamer, or other sensor-driven workloads.

Familiarity with compiled-model artifacts, quantized execution, tensor layouts, graph partitioning, memory planning, and compiler/runtime integration.

Experience with embedded Linux SDKs, production deployment, long-term runtime/API compatibility, or the security and isolation requirements of multi-process accelerator systems.