2 months ago
Remote, United StatesSenior / Staff+
H1B sponsor
Base Salary
$155k - $245k/yr
Responsibilities
- Port Deepgram speech models to non-NVIDIA and edge platforms with minimal model modification.
- Make serving-side decisions involving quantization, precision, operator substitution, graph rewrites, and architecture changes.
- Validate ports on real hardware using accuracy, latency, throughput, and memory benchmarks.
- Build repeatable model packaging, conversion, versioning, and automated delivery pipelines for edge targets.
- Collaborate with Embedded AI Engineers on custom kernels and adapt models to use them.
- Partner with platform and silicon vendors on runtimes, toolchains, model formats, and operator conventions.
- Feed edge deployment constraints back to Research and Impeller to improve future model portability.
- Take on additional edge production concerns such as automated deployment, model security, integrity, and fleet observability as the team grows.
Requirements
- Production experience deploying ML models to edge or non-NVIDIA hardware is required; cloud-only or GPU-only serving experience does not qualify alone.
- Working knowledge of quantization and precision tradeoffs, including INT8, FP16, mixed precision, and calibration.
- Experience with at least one edge or vendor inference runtime and its conversion toolchain, such as ONNX Runtime, TFLite, ExecuTorch, OpenVINO, Qualcomm AI Engine, or a vendor NPU SDK.
- Ability to read and rewrite model graphs, replace unsupported operators, and adjust architecture parameters while preserving accuracy.
- Strong Python and PyTorch skills plus production-quality practices for testing, reproducibility, and benchmarking.
- Experience building automation around model conversion and deployment.
- Clear communication and the ability to scope and drive model ports on unfamiliar platforms.
- Preferred experience includes speech, audio, or real-time streaming models; low-level kernel work; model security or integrity; Qualcomm, Apple, ARM, Intel, AMD, or custom NPU accelerators; and internal tooling for porting or 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.