Cortex Model
Build end-to-end ML pipelines from data validation through model deployment
What You Can Do
You can build production-ready ML pipelines that start with data validation and progress through model training, evaluation, and deployment. Cortex detects your ML stack, defines success metrics upfront, builds baseline models before optimization, and generates serving-ready artifacts—eliminating manual workflow steps and reducing time from data to deployed model.
Features
identifies your framework (scikit-learn, TensorFlow, PyTorch, XGBoost, LightGBM) and existing model artifacts
confirms prediction task, target metric, and baseline before any model training begins
prioritizes simple, production-proven models (logistic regression, gradient boosting) over complex architectures
schema checks, null handling, feature engineering, and train/test split automation
trains and evaluates multiple algorithms side-by-side with performance metrics
generates confusion matrices, ROC curves, feature importance, and error analysis
produces serialized models, prediction pipelines, and serving endpoint code
creates modular scripts (data_validation.py, train.py, evaluate.py, serve.py) for reproducibility
Example Output
Example 1: Classification Pipeline
✓ Detected: scikit-learn + pandas
✓ Task: Binary classification (churn prediction)
✓ Baseline metric: 72% accuracy (random)
✓ Model: Logistic Regression → 87% accuracy, F1: 0.84
✓ Alternative: XGBoost → 91% accuracy, F1: 0.89
✓ Deployment: FastAPI endpoint ready at localhost:8000/predict
Example 2: Regression Pipeline
✓ Detected: XGBoost + numpy
✓ Task: Price prediction (continuous output)
✓ Baseline metric: MAE $15,000 (mean estimate)
✓ Model: XGBoost → MAE $2,100, R²: 0.94
✓ Feature importance: square_footage (0.42), location (0.31), age (0.15)
✓ Serving code + model artifact generated
What's Included
- SKILL.md instruction file with ML pipeline workflow and environment detection logic:
- data_validation.py template: schema validation, null handling, feature engineering
- train.py template: baseline and advanced model training scripts
- evaluate.py template: metrics computation, confusion matrices, ROC curves, feature importance
- serve.py template: FastAPI/Flask endpoint code for model serving
- requirements.txt template: common ML dependencies (scikit-learn, XGBoost, TensorFlow, etc.)
Who It's For
- Data scientists building production ML pipelines without manual orchestration
- ML engineers automating model training and evaluation workflows
- Analytics engineers implementing predictive features in data platforms
- Backend engineers deploying machine learning models to serving endpoints
- Product teams rapidly prototyping classification and regression models
Best For
- Binary and multi-class classification tasks (churn, fraud, recommendation scoring)
- Regression models (price prediction, forecasting, resource estimation)
- Quick baseline establishment before hyperparameter tuning
- End-to-end pipeline generation from raw data to deployed model
- Model evaluation and diagnostics (feature importance, error analysis, metric comparison)
You might also like
This skill helps P&C underwriters evaluate risk profiles, recommend policy actions, and justify underwriting decisions through systematic analysis of applicant data, claims history, and exposure factors. You provide application details and historical information; Claude generates risk scores, compliance-checked recommendations, and documentation ready for underwriting files.
This skill analyzes medical histories, diagnoses, and clinical data to generate risk assessments and coverage recommendations for health insurance underwriting. You can streamline initial reviews, evaluate pre-existing conditions, and document coverage decisions with regulatory compliance checks built in. The skill helps you process medical documentation faster while maintaining consistency and accuracy in your risk evaluation process.
Transform complex commercial applications, loss histories, and operational data into professional risk assessments and coverage recommendations. You'll receive structured underwriting documentation with detailed risk analysis, exposure mapping, and loss pattern evaluation that meet regulatory standards. Streamline your underwriting workflow with data-driven, consistent assessments tailored to each client's unique risk profile.
You can systematically evaluate whether your insurance reserves are sufficient to cover incurred but not reported (IBNR) claims and identify reserve deficiencies. This skill analyzes claims development patterns, projects future claim emergence, and applies actuarial methods to assess reserve adequacy across product lines and underwriting years.
You can systematically review kriging assumptions, variogram interpretations, and search neighborhood designs to strengthen estimate credibility. This skill helps you diagnose bias and variance issues in block models, compare estimation methods with structured analysis, and map your methodology to JORC Table 1, NI 43-101, or CRIRSCO compliance requirements—turning complex geostatistical work into defensible technical documentation.
Upload your training logs, loss curves, and model architecture to Claude, which systematically diagnoses failures by analyzing convergence patterns, learning dynamics, and architectural inefficiencies. Claude recommends specific hyperparameter adjustments with scientific reasoning—learning rate schedules, regularization strengths, batch sizes, and architecture modifications—then guides you through iterative validation to confirm improvements.
Analyze your insurance portfolio to identify concentration risk across perils, geographies, and underwriting assumptions. Validate catastrophe model outputs against real exposure data and flagged data gaps. Generate concise executive risk summaries that highlight key vulnerabilities and inform underwriting decisions.
Transform unstructured property observations into comprehensive risk assessments with quantified hazard scores and prioritized control recommendations. Claude analyzes property details, identifies potential hazards across categories, calculates risk levels based on likelihood and impact, and delivers actionable remediation strategies tailored to your property type and use case.