Edge AI Engineer — Real‑Time, On‑Device Speech Models

Deepgram; United States; On-site; USD 140,000 - 190,000; Full time

Job summary

Deepgram is seeking an Embedded AI Engineer to bring high‑performance speech models to memory‑ and power‑constrained devices. You will optimize on‑device inference, write tight runtime code, and push models to run across mobile, embedded SoCs, microcontrollers, and dedicated accelerators.

In this role you’ll collaborate with silicon vendors and research teams to shape edge architectures, balance latency with accuracy, and deliver private, offline, real‑time voice experiences for consumer

Qualifications

• Experience delivering production systems on resource‑constrained hardware (embedded/mobile/edge).
• Proficient in C, C++, and/or Rust with performance‑critical code.
• Hands‑on experience with on‑device model optimization (quantization/pruning).

Responsibilities

• Run Deepgram models on embedded hardware across processors and accelerators.
• Optimize models for latency, memory, power, and thermal budgets.
• Write performance‑critical runtime code for embedded environments (C/C++/Rust).
• Integrate with edge runtimes and NPU/DSP toolchains for on‑device deployment.
• Build on‑device runtime plumbing: packaging, deployment pipelines, OTA updates.
• Benchmark and validate latency, accuracy, power, memory, and resource usage.
• Collaborate with silicon vendors for SDK integration and tuning.
• Partner with Research/Engine teams to steer edge‑friendly model designs.

Skills

C/C++/Rust

Embedded systems

Real-time systems

Communication skills

Tools

ONNX Runtime

TensorRT

TFLite

ExecuTorch

FreeRTOS

Zephyr

Job description

Deepgram is seeking an Embedded AI Engineer to bring high‑performance speech models to memory‑ and power‑constrained devices. You will optimize on‑device inference, write tight runtime code, and push models to run across mobile, embedded SoCs, microcontrollers, and dedicated accelerators.

In this role you’ll collaborate with silicon vendors and research teams to shape edge architectures, balance latency with accuracy, and deliver private, offline, real‑time voice experiences for consumer