https://drive.google.com/file/d/1-vUhhLsjTNU6qgViobVxys5v45srW-GL/view?usp=share_link

*Abstract*
Enterprise support ecosystems contain massive volumes of troubleshooting content, spanning
thousands of articles, dozens of product families, and unstandardized error codes. Customers often
arrive with only a cryptic diagnostic message—making it difficult for search engines or LLMs to route
them to the correct resolution article. In this talk, we present our work on building a Semantic
Error-Code Knowledge Graph (SEKG) and a set of KG-grounded AI Support Agents designed to
improve article discoverability, automate troubleshooting, and significantly enhance customer
self-service.
The Knowledge Graph unifies product metadata, error codes, symptoms, components, and support
articles into a single semantic model capable of disambiguating error messages across product
families.
Evaluation results show measurable
gains in retrieval accuracy, long-tail error-code coverage, and customer resolution effectiveness.

*Session Description*
Technical support interactions frequently begin with incomplete or ambiguous error codes—yet
customers expect instant answers. Traditional search struggles to connect these signals to the correct
articles, especially across large product ecosystems. This session presents a production-grade
approach using a Semantic Error-Code Knowledge Graph (SEKG) combined with KG-grounded AI
agents.
The session also shows how LLM-powered agents
leverage this graph to interpret customer queries and retrieve the closest relevant articles