Enterprise data analytics always runs into the same wall. "Explore internal data through natural language" is easy to say — but in practice, table names look like T_CUST_MST, and "net revenue" carries different calculation logic across departments. LLMs cannot understand business context from DDL alone.

DataNexus tackles this by combining an ontology-native NL2SQL engine, GraphRAG, and a Data Catalog. An open-source composite architecture that handles both unstructured documents and structured databases together.

This blog documents the process — including the dead ends. Architecture decisions, reasons behind technology choices, failed experiments alongside successful ones.