[12.08] Hayley — Inference: Fee Sub-agent enhancement - Fallback to KG

Fee Sub-Agent Fallback to Knowledge Graph

We've added automatic Knowledge Graph fallback when fee and utility data is missing from the primary collection.

This ensures complete fee information even for new properties with incomplete data → better response accuracy, fewer "I'm not sure" replies, and improved data coverage.

Read the full details here👇

What's New

• Two-tier data retrieval system queries primary collection first, then automatically falls back to Knowledge Graph nodes
• Single-query fallback retrieves 13 fee/utility types from Knowledge Graph in one roundtrip (application fee, amenity fee, trash fee, pet fee, utilities, etc.)
• Schema-compliant responses with fee_text fields populated from Knowledge Graph content and response metadata showing fallback source

Why This Matters: Intelligence Fabric Story

Before:

• Fee queries relied only on primary fees_and_utilities collection
• Missing or incomplete data returned empty responses
• New properties often had fee data in Knowledge Graph but not in primary collection
• "I'm not sure" responses even when data existed elsewhere

Now:

• Automatic fallback to Knowledge Graph when primary collection empty
• Single-query retrieval across 13 predefined KG paths
• Complete fee information from alternative sources
• Metadata tracking shows data source and fallback attempts

Business Impact

• Better response accuracy → retrieves fee data from Knowledge Graph when primary collection is empty
• Improved data coverage → new properties return complete fee information even before primary data ingestion.
• Fewer failed responses → automatic fallback prevents "I'm not sure" replies when data exists in KG.
• Low latency impact → less than 300ms added latency for fallback queries
• Full observability → response metadata includes fallback source and query timing for debugging

Built By

• Product Manager:Gul Jabeen
• Talha Khan

How to Use It

1. Hayley receives a fee or utility question from a prospect
2. System queries primary fees_and_utilities collection first
3. If primary collection returns empty, system automatically triggers fallback query to knowledge_graphs collection
4. Fallback retrieves data from 13 KG paths (application fee, amenity fee, trash fee, utilities, etc.) in single query
5. System transforms KG content into schema-compliant response with fee_text fields populated

Who Benefits

• Prospects: complete fee information even for newly onboarded properties
• Onboarding teams: immediate fee responses before primary data ingestion complete
• ML teams: better data coverage and fewer empty responses
• Support teams: reduced escalations from incomplete fee data