Our most recent community show explored how Reltio documentation has evolved from a website you visit into a structured corpus that AI tools can consume directly, and what that means for you. Asha K A and Luís Guedes explained how the team publishes content for machines, demonstrated three ways to use it, and shared the authoring principles behind it.
Why AI-Ready Content Matters
Documentation was historically built for one audience: human readers, who browse menus, search for keywords, and scan pages. . Now AI tools also retrieve, chunk, and use documentation to generate answers, whether through coding assistants like Cursor or Copilot, internal RAG pipelines, or support and enablement tools. Asha noted that scraping a website hurts retrieval in five ways:
-
Noise: Menus, footers, and navigation end up in the body text
-
Lost structure: Tables and code blocks lose their shape
-
Dead links: Cross-references become plain text
-
Silent staleness: Nothing signals that the content is out of date
-
Untraceable errors: Product changes and broken scrapes look identical
Publishing from the Source
Reltio documentation is authored as structured DITA XML in a content management system. The site at docs.reltio.com is one rendering of that source. The team now publishes a second, parallel rendering in Markdown, purpose-built for LLM grounding, chunking, and retrieval. Luís described the automated pipeline: it rebuilds the topic map from the source, converts content to Markdown, and uses smart hashing to update only the topics that changed. It produces two files:
-
docs.md: The full documentation corpus
-
index.md: A retrieval index with hierarchy paths, live URLs, keywords, short descriptions, and cross-references
Every file carries a generation timestamp, so freshness is easy to verify. The corpus refreshes every Wednesday and Friday.
Three Levels You Can Use Today
-
Level 1, the AI-Ready public corpus: The raw material for building your own AI layer. Download it with a simple curl command, no sign-in required, and drop it into a RAG pipeline, coding assistant, or large-context prompt
-
Level 2, the AI-Ready HOWTOs generator kit: An open generator that writes step-by-step guides grounded in official docs, available on github.com/reltio-ai. It works with Claude Code and refuses to write a guide if the docs offer too little source material on a topic
-
Level 3, Docs for You: A finished application with zero configuration. Sign in with your Reltio account, choose your role, goal, learning style, and experience level, and receive a tailored guide in about two minutes. Guides include linked sources, glossary definitions, diagrams where relevant, and a candid "questions this guide doesn't answer" section
Luís noted that Docs for You is not a chatbot, doesn't connect to your tenant, and shouldn't be given tenant data. The same content and refresh cadence sit underneath all three levels. The only difference is how much you want to build yourself.
What Makes Content AI-Ready
Asha shared five authoring principles the team applies across the documentation:
-
Self-contained pages: Each topic makes sense on its own when retrieved in isolation
-
Metadata on everything: Audience, version, and content type tags narrow the results
-
Meaningful links: Prerequisite and "see also" links tell machines how topics relate
-
Shallow paths: Flat structures with clear tags are easier to reach than deep hierarchies
-
Descriptive titles: Specific titles, like "token expiration rules," match questions better than "section 3.2"
Q&A Highlights
-
Domain-specific examples: Content isn't segregated by industry, so guides may use examples from other domains. Giving a specific task or use case works better than naming a domain, and the team invited attendees to share real scenarios for testing
-
SAP Joule: The tools aren't linked to Joule or any other product, and they use Reltio documentation portal content only
-
Role personas: Roles map to the approved product personas defined in the doc portal
-
How retrieval works: A first pass reads index.md to find relevant topics, code compiles the source snippets, and a second pass writes the guide from that grounded Markdown
Looking Ahead
All three deliverables are live, and the team welcomes feedback whenever an answer misses the mark.
#Featured
#CommunityWebinar
#AIML