Case study · AI
RAG knowledge corpus for AI agents
A 1.39M-row IBM i cross-reference database turned into 12,332 linked documents AI agents can search.
The client
An IBM i application team preparing for AI agents
The problem
Knowledge about how programs, files and fields relate lived in a ~236 MB cross-reference database that people and AI tools could not easily search.
What we built
- Converted ~1.39M rows across 13 tables into 12,332 linked markdown documents
- Added YAML frontmatter so each document carries structured metadata
- Packaged the corpus for MCP and retrieval-augmented generation (RAG)
The outcome
AI agents can answer questions about the IBM i application by retrieving the right documents instead of guessing.