Senior Firmware Engineer, Edge AI / NPU Runtime

Tacit•San Francisco, CA

2d•$150,000 - $200,000•Onsite

About The Position

We’re looking for a Senior Firmware Engineer, Edge AI / NPU Runtime to help architect, optimize, and ship next-generation neurotech hardware with production-grade on-device intelligence. You will own critical parts of the embedded AI stack, from realtime sensor acquisition through preprocessing, NPU/DSP-accelerated inference, postprocessing, telemetry, and product deployment. This is a hands-on role for someone who wants to work close to the hardware while shaping the intelligence users experience in the product. You’ll help define how models run on-device, how sensor data moves through the system, and how we meet tight latency, reliability, and power budgets in real-world use.

Requirements

5+ years of experience in embedded firmware, embedded systems, or edge ML systems.
Strong C/C++/Rust experience on resource-constrained embedded platforms.
Experience with RTOS-based systems such as FreeRTOS, Zephyr, ThreadX, or similar.
Experience deploying or optimizing ML inference on embedded targets, NPUs, DSPs, MCUs, or edge SoCs.
Strong understanding of realtime embedded systems, including DMA, interrupts, concurrency, memory management, and low-latency data movement.
Experience optimizing embedded systems for latency, memory footprint, throughput, and power consumption.
Hands-on debugging and bring-up experience across embedded hardware and firmware systems, with strong cross-functional communication across firmware, ML, electrical, software, and product teams.

Nice To Haves

Experience with embedded inference runtimes, deployment toolchains, or edge AI SoCs/accelerators such as TensorFlow Lite Micro, ONNX Runtime, CMSIS-NN, Qualcomm QNN/SNPE, ARM Ethos-U/Vela, TVM, ExecuTorch, Qualcomm, ARM, Cadence/Tensilica, Syntiant, Ambiq, Nordic, NXP, ST, Hailo, Google Edge TPU, or similar.
Experience with quantized inference, fixed-point math, SIMD/DSP optimization, accelerator programming, or model conversion workflows.
Experience with streaming or time-series ML workloads such as biosignals, sensor fusion, audio, gesture recognition, keyword spotting, or other realtime inference systems.
Experience shipping battery-powered consumer electronics, wearable, neurotech, AR/VR, robotics, camera, IoT, or other embedded AI products.

Responsibilities

Own deployment of ML models onto embedded targets using NPUs, DSPs, MCUs, or other hardware accelerators.
Integrate embedded inference runtimes, vendor NPU/DSP SDKs, and model deployment workflows into production firmware.
Optimize inference latency, memory footprint, throughput, power consumption, and accelerator utilization on production hardware.
Partner with ML teams on quantization, operator support, model architecture tradeoffs, calibration datasets, and accuracy/performance regressions.
Build realtime sensor-to-inference pipelines, including acquisition, timestamping, synchronization, preprocessing, feature extraction, model execution, and postprocessing.
Design low-latency data movement using DMA, interrupts, ring buffers, deterministic scheduling, and efficient memory layouts.
Support streaming inference patterns such as sliding windows, temporal models, event-driven execution, and continuous sensor processing.
Maintain inference quality and timing guarantees under real-world conditions such as sensor noise, clock drift, dropped samples, variable system load, and power-state transitions.
Optimize end-to-end energy per inference across sensing, preprocessing, model execution, postprocessing, and idle time.
Use low-power firmware techniques such as sleep states, duty cycling, subsystem power gating, clock scaling, batching/windowing, and dynamic power management.
Profile and improve power consumption across sensors, CPU, NPU/DSP, memory, and supporting firmware infrastructure.
Bring up and debug firmware across sensors, accelerators, power systems, embedded compute, and production hardware.
Use lab tools, traces, logs, telemetry, and instrumentation to root-cause complex embedded system issues.
Translate product and customer experience goals into concrete latency, reliability, responsiveness, and power targets.
Build diagnostics, validation hooks, and performance benchmarks to ensure reliable real-world edge inference behavior.