Principal Machine Learning Engineer

This range is provided by Circadia Health. Your actual pay will be based on your skills and experience — talk with your recruiter to learn more.

$/yr - $/yr

Annual Bonus and Stock options

Direct message the job poster from Circadia Health

Circadia Health is a growth-stage healthcare AI company on a mission to prevent avoidable hospitalizations and transform senior-care operations. Our Circadia Intelligence Platform combines:

• Contactless sensing that monitors respiration and motion with medical-grade accuracy
• Native predictive models that detect 85% of preventable adverse events several days in advance
• Enterprise integrations that operationalize predictions directly inside EHR, care-coordination, billing, and compliance workflows

Today, our technology touches 40,000+ post-acute patients daily across skilled-nursing, home-health, and home-care networks. We are backed by leading healthcare and AI investors and headquartered in El Segundo, CA.

Circadia’s core advantage is deep, production-grade predictive modeling on messy, high-stakes healthcare data—not demos, not dashboards, but models that materially change clinical and financial outcomes.

As a Principal ML Engineer, you will own the full lifecycle of native ML models—from feature engineering and model training to validation, monitoring, and continuous improvement in production. Your work will directly power risk stratification, early-warning systems, and agentic workflows used by clinicians and operators every day.

This is a role for someone who loves tabular data, time-series signals, causal nuance, and shipping models that actually work in the wild.

Build & Own Core Predictive Models

• Design, train, and iterate on XGBoost, Light
  
  GBM, Cat Boost, and other native ML models for risk prediction, classification, and regression
• Develop time-series and longitudinal models using vitals, motion, utilization, and claims-adjacent data
• Own feature pipelines spanning raw sensor outputs, clinical indicators, utilization patterns, and derived signals

End-to-End Model Lifecycle

• Define labeling strategies, handle missingness, censoring, and class imbalance
• Establish retraining cadences, drift detection, and performance guardrails

Clinical & Operational Rigor

• Partner with clinicians to ensure models are clinically interpretable, safe, and actionable
• Produce explainability artifacts (e.g., SHAP, feature attribution) suitable for audits, clinicians, and enterprise buyers
• Balance sensitivity/specificity trade-offs in real operational contexts (false positives matter)

Production & Platform Integration

• Collaborate with platform engineers to deploy models via APIs and batch pipelines
• Optimize inference latency and cost at scale
• Ensure models integrate cleanly into downstream agentic and workflow systems

Measurement & Outcomes

• Define and track real-world impact metrics (avoidable hospitalizations, LOS, cost reduction, staff efficiency)
• Run offline validation, shadow deployments, and post-deployment analyses
• Continuously improve models based on live outcomes, not just offline AUC

• 5–10 years of experience building and shipping native ML models in production environments
• Deep hands-on experience with XGBoost (required) and at least one of Light
  
  GBM / Cat Boost
• Strong foundation in statistics, ML fundamentals, and model evaluation
• Proven experience with tabular and/or time-series healthcare-like data (messy, sparse, biased, incomplete)
• Advanced Python skills; comfortable with Num Py, pandas, scikit-learn, and ML tooling
• Experience owning models end-to-end, not just experimentation notebooks
• Clear communicator who can explain model behavior and trade-offs to non-ML stakeholders
• High ownership mindset — you’ve carried models through failures, audits, and real-world edge cases

• Experience in healthcare, insurance, fintech, or other regulated, high-signal domains
• Familiarity with survival analysis, hazard models, or early-warning systems
• Experience with sensor data, physiological signals, or remote monitoring
• Comfort working alongside LLM-based systems (even if you don’t build…