Edge ML Engineer: CPU-Only Video Analytics on Routers

New York Technology Partners; Irving (TX); On-site; USD 120,000 - 180,000; Full time

Job summary

New York Technology Partners is seeking an experienced engineer to port GPU-based video analytics models to CPU-only router targets and optimize the on-device inference pipeline. The role includes building a containerized architecture with dynamic cloud-driven model loading and tuning accuracy versus performance on ARM/MIPS router hardware.

You will integrate with Cradlepoint OS and PrplOS environments, benchmark latency, and iterate on detection accuracy on constrained hardware, ensuring memory

Qualifications

• 4+ years in embedded systems or edge ML deployment.
• Experience with containerization (Docker, LXC) on constrained devices.
• Video analytics / computer vision (YOLO variants, object detection pipelines).
• Python + C/C++ on Linux embedded targets.
• Cross-compilation, profiling, and memory optimization.

Responsibilities

• Port GPU-based video analytics models (object detection, classification) to CPU-only router targets.
• Optimize inference pipeline to stay under 100MB memory footprint using SLMs.
• Build containerized architecture with dynamic cloud-driven model loading.
• Tune accuracy/performance tradeoffs on ARM/MIPS router hardware.
• Integrate with Cradlepoint OS and PrplOS environments.
• Benchmark and iterate on detection accuracy vs. latency on constrained hardware.

Skills

Embedded systems

Edge ML deployment

Containerization

Video analytics

Python

C/C++

Cross-compilation

Profiling

Memory optimization

Tools

Docker

LXC

Job description

New York Technology Partners is seeking an experienced engineer to port GPU-based video analytics models to CPU-only router targets and optimize the on-device inference pipeline. The role includes building a containerized architecture with dynamic cloud-driven model loading and tuning accuracy versus performance on ARM/MIPS router hardware.

You will integrate with Cradlepoint OS and PrplOS environments, benchmark latency, and iterate on detection accuracy on constrained hardware, ensuring memory