Applied ML Engineer - Edge Devices
Deepgram · USA - Remote
About this role
## Applied ML Engineer — Edge Devices ### Company Overview Deepgram is the leading platform powering the Voice AI economy, offering real-time APIs for speech-to-text (STT), text-to-speech (TTS), and production-grade voice agents at scale. With voice-native foundation models available via cloud APIs or self-hosted/on-premises software, Deepgram delivers high accuracy, low latency, and cost efficiency. ### About the Role Deepgram’s speech models run at scale on NVIDIA GPUs, but customers increasingly need them on non-NVIDIA accelerators, edge servers, and embedded platforms with their own inference runtimes, operator sets, and constraints. As an **Applied ML Engineer** on the **Partner Platform Engineering** team, you’ll adapt Deepgram models to run efficiently on target edge platforms—minimizing model changes while fitting the platform’s existing kernel/runtime paradigm. This is **not** a research role and **not** a cloud-serving role; it’s applied ML for edge deployment. ### What You’ll Do - **Port Deepgram speech models** to non-NVIDIA and edge platforms by adapting model structure/parameters to match the target operator set, runtime, and kernels. - Own **serving-side model decisions** for edge targets, including quantization/precision, operator substitution, graph rewrites, and architecture tweaks. - **Validate on real hardware** with accuracy, latency, throughput, and memory benchmarks; catch regressions before customers do. - Build the **deployment path** for edge targets: packaging, conversion pipelines, versioning, and automated delivery for repeatable shipping. - Collaborate with **Embedded AI Engineers** when custom kernels are needed—specify model requirements and integrate with delivered kernels. - Partner with **platform and silicon vendors** on runtimes/toolchains to produce working Deepgram deployments. - Feed edge constraints back to **Research and Impeller** to make future models easier to port. - As the team grows, take on additional edge production needs like deployment automation, model security/integrity, and fleet observability. ### You’ll Love This Role If You - Have fought to get models running on hardware not built for them—and want to do it across many platforms. - Prefer changing the model to fit the hardware (and know when to do otherwise). - Care about device performance metrics, not just notebook results. - Enjoy bridging the gap between kernel-writing and model-training teams. - Want to ship to customers, not publish. ### It’s Important To Us That You Have - **Hands-on experience deploying ML models to edge or non-NVIDIA hardware in production** (required). Cloud-only or GPU-only serving experience alone does not qualify. - Working knowledge of **quantization and precision tradeoffs** (INT8, FP16, mixed precision, calibration) and their impact on accuracy/latency. - Experience with at least one **edge/vendor inference runtime** and its conversion toolchain (e.g., ONNX Runtime, TFLite, ExecuTorch, OpenVINO, Qualcomm AI Engine, or a vendor NPU SDK). - Ability to **modify models to fit a platform**: read/rewrite model graphs, swap unsupported operators, adjust architecture parameters without breaking accuracy. - Strong **Python and PyTorch** with production-quality engineering habits (tests, reproducibility, benchmarks). - Comfort building **automation** around model conversion and deployment. - A builder mindset and clear communication to scope ports on unfamiliar platforms and drive measurable results.
Listing freshness
CronJobs last confirmed this listing 1h ago. If its source stops confirming the opening for seven days, this page is removed from active inventory.