Architecture overview
Guides Knowledge AI is a platform for building AI-powered knowledge assistants from your existing documentation. The diagram below shows the end-to-end architecture.
Guides Knowledge AI platform architecture
Content flows into the platform from two sources:
- AEM Guides — native DITA content with rich metadata (prolog, shortdesc, navigation titles) is indexed directly into the knowledge graph.
- External systems — content in HTML, Markdown, PDF, or DOCX formats is ingested through AI-assisted conversion pipelines that transform it into structured DITA before indexing. Metadata such as taxonomy and ontology information is preserved and used for entity extraction and retrieval.
Each domain (D1, D2, ... Dn) is an isolated knowledge container with its own knowledge graph, model configuration, indexed documents, and chat history. Domains let you organize content by product, team, or use case while keeping configurations and analytics separate.
Users and applications access domain knowledge through multiple channels:
- MCP — connect AI-powered developer tools (Cursor, Claude Desktop) directly to a domain. Also powers Adobe Brand Concierge for brand-aligned AI assistance.
- API — programmatic access via API keys for custom integrations and workflows. Includes Slack integration to bring Q&A into Slack channels for team collaboration.
- Chatbot widget — an embeddable widget for websites with configurable appearance and behavior.
The platform provides a broad set of tools for managing, optimizing, and monitoring your knowledge assistant:
- Analytics & Reports — usage trends, LLM latency, answer success rates, cache hit rates, and top users.
- Feedback — human feedback (thumbs-up/down) and LLM-as-a-Judge automated scoring, with side-by-side comparison.
- Prompt Management — customize answer generation and entity extraction prompts to tailor chatbot behavior.
- Golden Dataset & Evaluations — upload question-answer datasets, run experiments, and measure answer quality over time.
- A/B Testing — change a model, prompt, or retrieval parameter, then re-run an evaluation experiment against the same dataset to compare results side by side and find the best configuration for your domain.
- Content Optimization — AI quality scoring of indexed documents with per-criterion breakdowns and actionable suggestions.
- User Management & Security — custom roles with granular org-scoped and domain-scoped permissions, API key management.
- BYOM — Bring Your Own Model support for Azure OpenAI, WatsonX, and Gemini, with custom LLM, embedding, and evaluation model configurations.
FAQ
What is Guides Knowledge AI and what does the architecture overview describe?
Guides Knowledge AI is a platform for building AI-powered knowledge assistants from your existing documentation. The architecture overview describes the end-to-end flow from content supply, into isolated knowledge domains, out through multiple consumption channels, supported by platform capabilities for management, optimization, and monitoring.
How does content get into Guides Knowledge AI from AEM Guides and external systems?
From AEM Guides, native DITA content with rich metadata (such as prolog, shortdesc, and navigation titles) is indexed directly into the knowledge graph. From external systems, HTML, Markdown, PDF, or DOCX content is ingested through AI-assisted conversion pipelines that transform it into structured DITA before indexing. Taxonomy and ontology metadata is preserved and used for entity extraction and retrieval.
What are knowledge domains and why are they used?
A knowledge domain is an isolated knowledge container with its own knowledge graph, model configuration, indexed documents, and chat history. Domains let you organize content by product, team, or use case while keeping configurations and analytics separate.
What consumption channels can users and applications use to access domain knowledge?
Domain knowledge can be accessed through MCP, API, and a chatbot widget. MCP connects AI-powered developer tools (like Cursor and Claude Desktop) directly to a domain and also powers Adobe Brand Concierge. The API provides programmatic access via API keys and includes a Slack integration, while the chatbot widget is an embeddable website widget with configurable appearance and behavior.
What platform capabilities are available to manage, optimize, and monitor a knowledge assistant?
The platform includes Analytics & Reports, Feedback (human and LLM-as-a-Judge scoring), Prompt Management, Golden Dataset & Evaluations, and A/B Testing to compare configurations against the same dataset. It also provides Content Optimization with AI quality scoring and suggestions, User Management & Security with granular permissions and API key management, and BYOM support for Azure OpenAI, WatsonX, and Gemini with configurable LLM, embedding, and evaluation models.