Perception / ML Engineer

Open People Network (OPN); Toronto; On-site; CAD 120,000 - 170,000; Full time

Job summary

Osprey Systems seeks a skilled ML/edge AI engineer to fuse multi-sensor data into real-time, device-local inferences with sub-second latency. You will design a robust fusion engine, develop target classification under challenging conditions, and implement trajectory analysis on embedded GPUs.

You will work on field-collected data, labeling workflows, and model retraining, with a bias toward real-world validation. Collaboration across small teams and future hiring will shape the role.

Qualifications

• 4+ years building and shipping ML systems, including perception or tracking.
• Strong classical estimation alongside deep learning: filtering, data association, probabilistic reasoning.
• Edge or embedded deployment experience with real latency and memory constraints.
• Proficiency in Python and C++ for production-grade systems.
• Comfort with self-collected data and bias toward field validation over benchmarks.

Responsibilities

• Design and build the multi-sensor fusion engine that cross-validates detections into unified tracks with calibrated confidence.
• Develop classification models that separate targets from clutter and benign activity.
• Build behavioral inference: trajectory analysis, pattern recognition, and assessment.
• Deploy and optimize models for edge inference on embedded GPU hardware within strict latency budgets.
• Build field collection, labeling, dataset versioning, and retraining workflows.
• Handle low light, snow, fog, acoustic and RF noise as normal operating conditions.
• Help shape engineering culture, tooling, and hiring as the team grows.

Skills

ML systems

Edge computing

Field validation

Low-latency design

Python

C++

Tools

Python

C++

TensorRT

ONNX

CUDA

Job description

Osprey Systems is building Canada's sovereign counter-drone capability. Our distributed detection and interception system is designed for the environments where everything else fails: extreme cold, GPS-denied, and network-degraded conditions. We serve critical infrastructure operators and defense customers with a single product line, built in Canada.

We are a small founding team backed by a clear funding pathway through Canadian defense innovation programs and paid civilian deployments.

The role

Our systems sense the world through several independent modalities at once. Your job is fusing them into one trustworthy answer, computed on the device itself, in well under a second, without cloud, satellite navigation, or a reliable network. False positives are what kill sensing products. Your work is the cure.

What you will do

• Design and build the multi-sensor fusion engine that cross-validates independent detections into unified tracks with calibrated confidence.
• Develop classification models that separate targets of interest from benign activity and environmental clutter.
• Build behavioral inference on top of classification: trajectory analysis, pattern recognition, and assessment.
• Deploy and optimize models for edge inference on embedded GPU hardware under strict latency budgets.
• Build field collection, labeling, dataset versioning, and retraining workflows.
• Handle low light, snow, fog, acoustic noise, and RF noise as normal operating conditions.
• Help shape engineering culture, tooling, and hiring as the team grows.

What we are looking for

• 4+ years building and shipping ML systems, including perception, tracking, or multi-sensor problems.
• Strong classical estimation alongside deep learning: filtering, data association, probabilistic reasoning.
• Edge or embedded deployment experience with real latency and memory constraints.
• Python and C++ proficiency.
• Comfort with small, messy, self-collected datasets.
• Bias toward field validation over benchmark metrics.

Nice to have: Classification in cluttered or adversarial sensor environments; multi-object tracking; robotics or autonomous-systems perception; TensorRT, ONNX or CUDA depth; early-stage startup experience.