About the role:

Safety AI builds the ML and computer vision systems behind Samsara's AI dash cameras which enable real-time driver alerts, risk signals, and coaching insights running on millions of edge devices and the cloud.

This role owns what happens after training: building the resilient, low-latency ML backend systems that turn static model artifacts into high-throughput, cloud-scale safety features.

You will partner closely with applied scientists, firmware and full-stack engineers, and product managers and you will build the ML APIs, data pipelines, and evaluation infrastructure that let Safety AI models run efficiently at fleet scale, closing the loop from initial integration through rollout monitoring and iteration to a trustworthy, customer-facing signal. 

This is ML engineering where the stakes are real: rare, high-consequence events, millions of vehicles, and a product where "it works" means someone got home safely. Kindly refer to this video.

This is a remote role open to candidates residing in the US or Canada.

Technical Charter and Impact:

• Own the cloud-side path from model artifact to production system for Safety AI's ML applications.
• Establish practical standards for productionizing models — how they're served, evaluated, versioned, and monitored once they leave applied science.
• Set a high bar for reliability: rigorous evaluation, measurable rollout health, and systems that degrade predictably rather than silently.
• Act as a technical partner to applied scientists, helping translate research outputs into systems that are debuggable, scalable, and cost-efficient in production.

In this role, you will: 

• Design Production ML APIs: Architect and maintain reliable, low-latency APIs to integrate Safety AI model outputs directly into cloud applications.
• Build Data Flywheels: Construct scalable pipelines to power continuous model iteration, backtesting, shadow and online evaluation, enabling fast and safe deployments.
• Optimize & Serve Artifacts: Productionize model artifacts handed off by applied scientists, optimizing serving logic and fine-tuning models for platform-specific workloads.
• Petabyte-Scale Operations: Process high-volume camera and sensor telematics data to support model execution, backtesting, and automated dataset curation.
• Monitor & Maintain Rollout Health: Build systems to track model drift, precision/recall, and latency regressions in production, ensuring predictable failure modes and closed-loop data feedback.
• Cross-Functional System Integration: Partner with firmware and platform teams to optimize edge-to-cloud model execution, balancing latency, throughput, and infrastructure cost.
• Product Collaboration: Work with product managers to translate safety requirements into scalable technical architectures.
• Champion, role model, and embed Samsara’s cultural principles (Focus on Customer Success, Build for the Long Term, Adopt a Growth Mindset, Be Inclusive, Win as a Team) as we scale globally and across new offices.

Minimum requirements for the role:

• 6+ years of experience as a Machine Learning Engineer or similar role, with a track record of shipping models in production.
• Strong proficiency in one or more common languages (e.g., C++, Golang, Java, Python, Scala).
• Proficiency with common ML tools (e.g. Ray/Ray Serve, MLflow, Grafana, Pytorch, Spark, etc).
• Experience deploying and iteratively refining models using real customer feedback loops.
• Comfort with full-stack/backend development — you understand the data structures and dependencies underneath your models.
• BS or MS in Computer Science or a related quantitative field.

An ideal candidate also has:

• Ph.D. in Computer Science or a quantitative discipline (e.g., Applied Math, Physics, Statistics).
• Experience with containerization (Docker, Kubernetes), CI/CD pipelines, and infrastructure-as-code frameworks.
• Experience deploying and managing ML applications in cloud environments (AWS/GCP/Azure), including cloud-based storage, processing, and inference.
• Experience shipping end-to-end ML applications from artifact to production, ideally in safety-critical or high-scale domains.
• Expertise optimizing distributed model training with GPUs.