Agentic Feature Interface

The agentic feature interface enables AI agents to autonomously discover, query, and use features from the feature store through natural language interactions. This bridges the gap between conversational AI and structured feature management.

Overview

The AgenticFeatureInterface allows agents to:

• Discover features by describing what they need in natural language
• Query features for specific entities without knowing the feature view names
• Compose feature sets for training by describing the prediction task

Agent-Driven Feature Discovery

Agents can search for relevant features using semantic descriptions:

from src.features.agentic_feature_interface import AgenticFeatureInterface
 
interface = AgenticFeatureInterface(store=unified_feature_store)
 
# Agent asks: "I need customer purchase behavior features"
features = await interface.discover_features(
    tenant_id="acme-corp",
    description="customer purchase behavior and spending patterns",
    entity_type="customer",
)

Feature Lookup by Description

# Agent asks: "What is the average order value for customer C123?"
result = await interface.lookup_by_description(
    tenant_id="acme-corp",
    entity_id="C123",
    feature_description="average order value",
)

Source Files