4 months ago
Remote, United StatesSenior / Staff+
H1B sponsor
Base Salary
$219k - $274k/yr
Responsibilities
- Write and optimize custom kernels and operators in C, C++, Rust, and platform assembly or intrinsics for non-NVIDIA accelerators, embedded SoCs, DSPs, and NPUs.
- Optimize models for target hardware through quantization, operator fusion, memory layout, and architecture-specific compilation while meeting latency, memory, power, and thermal budgets.
- Integrate vendor NPU/DSP toolchains and edge inference runtimes, extending them with custom operators where needed.
- Deliver reusable kernels and runtime components with documented interfaces and constraints for Applied ML Engineers.
- Build performance-critical runtime code for embedded Linux, bare-metal, and RTOS targets.
- Establish platform benchmarking and validation for latency, accuracy, power, memory footprint, and utilization, and prevent regressions.
- Partner with silicon and platform vendors on SDK integration and low-level performance tuning for new chipsets and reference platforms.
- Provide hardware-constraint feedback to Applied ML and Research teams.
Requirements
- Production experience with resource-constrained hardware such as embedded systems, mobile, edge AI, or low-power devices.
- Strong proficiency in C, C++, and/or Rust, with experience writing performance-critical code for constrained environments.
- Hands-on experience optimizing models for on-device deployment, including quantization, pruning, knowledge distillation, or architecture-specific compilation.
- Familiarity with edge inference runtimes such as ONNX Runtime, TensorRT, TFLite, or ExecuTorch and/or vendor-specific NPU/DSP toolchains.
- Strong understanding of CPU, GPU, NPU, and DSP architectures, memory hierarchies, fixed-point/integer arithmetic, and power management.
- Experience with bare-metal or RTOS environments such as FreeRTOS or Zephyr, embedded Linux, microcontrollers, or edge SoC development.
- Strong communication skills and the ability to scope ambiguous optimization problems, drive measurable results, and explain tradeoffs.
- Preferred: experience with embedded real-time audio processing, DSP pipelines, audio codec optimization, wake-word or always-on listening, or streaming inference.
- Preferred: experience with custom quantization, mixed-precision inference, neural architecture search, hardware benchmarking, consumer AI products, model compilation toolchains, or secure on-device deployment.
Categories
About Deepgram
Deepgram builds Voice AI infrastructure for developers, offering speech-to-text, text-to-speech, and voice agent tooling through real-time APIs and deployable on-prem software. Privately held and founded in 2015, the company is headquartered in San Francisco. Customers such as Twilio, Cloudflare, and Jack in the Box use its models to power production voice features; products include Nova-3 STT, Aura TTS, and a Voice Agent API.