Embedded Computer Vision Engineer (Edge Inference)

Rapsodo is a global sports technology company with offices in Singapore, the United States, Türkiye, Japan, and Malaysia. We build data-driven, portable, and easy-to-use sports analytics products that help athletes at every level understand and improve their performance. From Major League Baseball pitchers to professional golfers, our technology is trusted by athletes and coaches worldwide — from youth development to elite competition. Our products deliver real-time, actionable insights that directly impact performance. We operate at the intersection of hardware and software, designing and building integrated systems with a strong emphasis on accuracy, reliability, and performance. Rapsodo is a globally distributed team, working across regions, cultures, and time zones. What enables us to operate effectively is not just technical capability, but how we work — with ownership, clarity, and disciplined execution. We are building computer-vision capabilities on Linux-based edge devices. This role owns the embedded software that takes models from “works on a workstation” to “runs reliably, efficiently, and measurably fast on-device.” You will develop and optimize inference pipelines, integrate vendor runtimes on NPUs/MPUs, and work close to the Linux kernel when needed (performance, memory, I/O, and driver interactions). What you will do Build and maintain production-grade embedded software for on-device computer vision inference (camera ingest, preprocessing, inference, postprocessing, telemetry) primarily in C++, with Rust as an option where appropriate. Integrate and run deep learning models using edge runtimes/toolchains (e.g., TensorRT, TFLite, OpenVINO, ONNX Runtime, vendor SDKs for NPUs/MPUs). Profile and optimize end-to-end performance: latency, throughput, memory footprint, power, and thermal constraints. Implement deployment-oriented model optimizations when needed (quantization workflows, operator compatibility fixes, graph optimizations, runtime-specific conversion). Work on Linux-based embedded platforms: cross-compilation, build systems, packaging, and reliable field deployment. Debug complex system issues across the stack: kernel/user-space boundaries, driver/I/O bottlenecks, memory contention, and multi-threaded performance. Collaborate with model/CV stakeholders to ensure models are edge-ready (I/O specs, accuracy vs. performance tradeoffs, validation on target hardware). Establish and uphold engineering standards: code quality, test strategy, CI, performance benchmarks, and observability on-device. Required qualifications Master’s degree or PhD in a relevant field (Computer Vision, Machine Learning/Deep Learning, Electrical/Computer Engineering, Computer Science, or related). Min 12 years professional experience in embedded software development, with significant time shipping Linux-based products. Strong expertise in C++ (modern C++11/14/17); Rust experience is a plus (or willingness to use Rust where it benefits reliability/performance). Strong Linux systems knowledge, including at least some of: kernel fundamentals, device I/O, scheduling, memory behavior, and profiling/debugging tooling (e.g., perf, ftrace, eBPF). Working knowledge of computer vision and deep learning inference concepts (pipelines, tensors, common CV tasks, latency/accuracy tradeoffs). You do not need to be a model developer/researcher, but must be fluent in deploying and running models. Experience optimizing inference for edge hardware (NPUs/MPUs/GPUs/accelerators), including quantization and runtime constraints. Preferred qualifications Camera stacks and media pipelines (V4L2, GStreamer, ISP integration). Embedded build and deployment toolchains (Yocto/Buildroot, CMake/Bazel). Hardware-aware optimization experience (ARM, NEON/SIMD). Experience with vendor-specific NPU SDKs and quantization toolchains (e.g., Rockchip RKNN, Qualcomm SNPE/QNN, MediaTek, Intel Movidius, etc.). OTA, reliability, and embedded security practices (watchdogs, crash dumps, secure boot). AI coding tools Comfortable using modern AI-assisted development tools (e.g., code completion, refactoring, test generation) while maintaining strong engineering judgment, code review discipline, and security awareness.