kailash-ml-protocols

Frozen interface contracts for the Kailash ML ecosystem. This thin package (~160 lines, zero dependencies) breaks the circular dependency between kailash-ml and kailash-kaizen.

Install

pip install kailash-ml-protocols

No dependencies required -- Python 3.10+ stdlib only.

Protocols

MLToolProtocol

Tools that Kaizen agents call via MCP to access ML capabilities.

• predict(model_name, features, *, options=None) -- Single-record prediction
• get_metrics(model_name, version=None, *, options=None) -- Model metrics
• get_model_info(model_name, *, options=None) -- Model metadata

AgentInfusionProtocol

Protocol for agent-augmented engine methods.

• suggest_model(data_profile, task_type, *, options=None) -- Suggest model families
• suggest_features(data_profile, existing_features, *, options=None) -- Suggest feature engineering
• interpret_results(experiment_results, *, options=None) -- Interpret experiment results
• interpret_drift(drift_report, *, options=None) -- Interpret drift report

Schemas

• FeatureField -- Single feature column definition
• FeatureSchema -- Schema for a feature set with to_dict() / from_dict() round-trip
• ModelSignature -- Input/output schema for a trained model
• MetricSpec -- A single evaluation metric with its value

License

Apache 2.0 -- Terrene Foundation