This content is currently WIP. Diagrams, content, and structure are subject to change.
This section covers surfacing Jupyter notebooks, providing tutorials on how C3 uses machine learning models, and how to set up classes and what they connect to.

​
AI & ML Capabilities for Data Scientists

The C3 Agentic AI Platform provides a comprehensive suite of tools and capabilities for data scientists and ML engineers to build, train, deploy, and monitor sophisticated machine learning models. These capabilities are designed to accelerate the ML lifecycle while ensuring enterprise-grade reliability and scalability.

​
End-to-End ML Lifecycle Support

From data preparation to model deployment and monitoring, the C3 Agentic AI Platform provides a complete toolkit for managing the entire ML lifecycle. This integrated approach eliminates the need to stitch together multiple tools and platforms, reducing complexity and accelerating time to value.

​
Feature Engineering & Management

The platform’s Feature Store provides robust capabilities for creating, managing, and using features in your ML models:
  • Feature Creation: Easily create features from C3 Metrics or custom Python functions
  • Materialization Options: Support for both full and incremental materialization to optimize performance
  • Feature Sets: Create, manage, and evaluate feature sets with point-in-time joins for accurate historical analysis
  • Reproducibility: Snapshot capabilities ensure reproducible results across experiments
  • Flexibility: Global materialization disable option for testing and development

​
Flexible Model Development

Build models using your preferred frameworks and libraries, with support for the most popular ML tools:
  • DAG-based Pipelines: Create complex ML workflows using MlPipeline with directed acyclic graph (DAG) architecture
  • Wide Library Support:
    • scikit-learn
    • TensorFlow
    • Keras
    • PyTorch
    • XGBoost
    • LightGBM
    • spaCy
    • Hugging Face Transformers
    • GluonTS
    • sktime
    • catboost
    • Custom Python/R/Java models

​
Advanced ML Capabilities

Take advantage of sophisticated ML techniques and optimizations:
  • Custom Pipeline Components:
    • MlLambdaPipe for stateless function integration
    • MlDynamicPipe for stateful class integration
    • MlTemplate for reusable pipeline patterns
  • Hyperparameter Optimization: Scale your HPO with Optuna, supporting:
    • Random search
    • Grid search
    • Tree-structured Parzen Estimator (TPE) search
  • Model Interpretability: Understand your models with:
    • ShapInterpreter with KernelExplainer and TreeExplainer
    • Background data support for more accurate interpretations
  • Distributed Training: Scale model training with PyTorch Distributed Data Parallelism

​
Enterprise-Grade Model Deployment

Deploy models with confidence using the Model Deployment Framework (MDF):
  • Lifecycle Management: Deploy, retrain, and retire models with built-in workflows
  • Flexible Routing: Direct predictions to the right models with relation-based and rule-based routing
  • Output Handling: Store and process predictions with database and CSV output handlers
  • Resilience: Automated recovery for ML jobs ensures reliability
  • Model Registry: Comprehensive registration, versioning, search, and approval workflows