Skip to main content
On this page

Intelligent & semantic search

Property management operations accumulate vast volumes of unstructured information: historical assembly minutes, maintenance logs, lease agreements, and service invoices. The etg24 platform makes this information instantly discoverable — not just through exact keyword matching, but by semantic intent.

Unified search across three dimensions

  1. Structured entity search: Perform filtered, indexed lookups across core relational entities: properties, units, tenant/owner contacts, and contracts.
  2. Deep document retrieval: PDF and image uploads are OCR-processed, chunked, and indexed with embeddings for hybrid retrieval (combining lexical BM25 and vector search).
  3. Knowledge base grounding: The official etg24 user manual is indexed in the same vector space. AI agents can search the knowledge base via meta.docs.search to resolve operational questions without hallucination.

API and MCP integration

Search capabilities are accessible across both protocols:

  • REST API: Query endpoints such as GET /propertymanagement/v2/query/search provide structured filtering, pagination, and score ranking. See the Property Management v2 reference.
  • MCP Server: Autonomous agents leverage specialized meta-tools:
    • meta.tools.search — Finds the optimal domain tools for a given natural language intent.
    • meta.docs.search — Performs semantic vector retrieval across documents and product documentation.
    • meta.thesaurus.search — Maps industry-specific German and English real estate terminology (e.g. WEG terminology).

Production example: Context-grounded support

When an AI agent is asked: “Are there any reported water damages for building Schillerstraße 12?”

  1. It searches structured records to resolve the property ID for Schillerstraße 12.
  2. It executes a semantic search over recent maintenance tickets and communication notes matching “pipe burst, leakage, moisture”.
  3. It returns an accurate, context-grounded summary referencing exact document dates and ticket IDs — without relying on ungrounded model assumptions.