Deepgram
Embedded AI Engineer, On-Device Models
$219K–$274K
USA - Remote · Remote · Full-time · Posted Jul 7, 2026
About the job
Builds and optimizes low-level kernels, operators, and runtime components to deploy Deepgram speech models on constrained, non-NVIDIA and embedded hardware.
What you'll do
- Write custom kernels and operators and optimize models for device-specific latency, memory, power, and thermal constraints.
- Integrate vendor toolchains and edge inference runtimes; build and deliver reusable embedded runtime components.
- Benchmark and validate performance, accuracy, power, memory use, and utilization across platforms, and work with vendors on hardware integration.
What they ask for
- Production systems experience on resource-constrained hardware.
- Strong proficiency in C, C++, and/or Rust, including performance-critical code for constrained environments.
- Hands-on on-device model optimization experience, such as quantization, pruning, knowledge distillation, or architecture-specific compilation.
- Understanding of hardware-software interaction and experience with embedded Linux, bare-metal, RTOS, or microcontroller and edge SoC development.
Nice to have
- Real-time audio processing experience on embedded platforms.
- Experience with edge ML optimization, hardware evaluation and benchmarking, or shipping consumer AI features.
Pay
$219.3K – $274.1K USD base salary; 10% Annual Bonus
Tools
C, C++, Rust, ONNX Runtime, TensorRT, TFLite, ExecuTorch, FreeRTOS
AI focus
Optimizes and deploys speech AI models on embedded and edge hardware, including custom kernels, quantization, and device-specific execution.
Model-written summary · the original posting is authoritative