Semantic CMDB Search Tool
A prototype that enables natural language search across configuration and dependency data to surface relevant systems, relationships, and evidence.
Overview
This prototype applies semantic embeddings and a lightweight retrieval layer to a configuration management database. It accepts natural language queries and returns ranked configuration items and dependency paths with contextual evidence.
Challenge
Finding the right configuration or dependency information was slow and error prone. Teams spent time manually searching multiple systems and often missed the context needed for rationalization and impact analysis.
Approach
I created a pipeline to normalize configuration text, encoded items with embeddings, and built a query interface that rewrites user queries into embeddings before performing similarity search. The UI surfaces results with supporting attributes and relationship links for traceability.
Results
Prototype testing reduced time to identify candidate systems by a large margin in user tests and increased stakeholder confidence by including explanatory evidence with each result. The approach proved valuable for application discovery and portfolio rationalization use cases.