Knowledge Base

Overview

The Knowledge Base is the layer that gives Denzing Agents true business understanding. It captures your organization’s terminology, rules, and context so answers reflect how your business actually operates. 

Without this layer, an Agent can read data but cannot fully interpret it. This can lead to answers that are technically correct but misaligned with business meaning. A well-structured Knowledge Base ensures that every response is accurate, consistent, and grounded in your domain. 

When a user asks a question, the Agent retrieves relevant context from the Knowledge Base before generating a response. 

How the Agent Reads the Knowledge Base

Denzing does not read documents sequentially. It retrieves only the most relevant pieces of information based on the user’s question. 

Chunking and Retrieval

• Documents are split into smaller units called chunks 
• Each chunk is converted into a searchable representation 
• For every query, the system retrieves a small number of the most relevant chunks 
• These retrieved chunks form the context for the final answer 

Each chunk must be self-contained. The Agent may only see a few chunks at a time, so every unit should clearly communicate its meaning without relying on surrounding text. 

Document Structure

Knowledge Base documents follow a strict two-level hierarchy:

• Section headings (##) define a broad topic 
• Chunk headings (###) define individual rules, definitions, or instructions 

## Section Heading       ← Broad topic area  
### Chunk Heading        ← One specific rule, definition, or instruction

All usable knowledge must exist inside clearly defined chunks. Any text outside of this structure may be ignored or misinterpreted. 

Writing Sections

Section headings should clearly describe the topic they cover. 

Each section should:

• Use precise, descriptive language 
• Reflect real business concepts or domains 
• Be followed by one clear sentence describing the scope 

Avoid vague or generic headings. Clear structure improves retrieval accuracy and ensures the Agent can locate the right information. 

Instead of	Use this
## Overview	## Patient Metric Definitions
## Notes	## Date and Season Handling Rules
## Miscellaneous	## Refund Eligibility and Order Status Rules

Writing Chunks

Chunks are the core building blocks of the Knowledge Base. 

Best Practices

• One concept per chunk: Each chunk should define a single rule, concept, or instruction 
• Start with a clear summary: Begin with one sentence that explains what the rule is and why it matters 
• Use specific headings: Headings should reflect real business terms that users are likely to search for 
• Repeat key terms naturally: Important terms should appear in both the heading and the body 
• Explain before showing code: Always describe what a query or formula does before including it 

Standard Structure

A clear sentence describing the rule or concept. 
Additional context explaining when it applies and why it matters. 
Optional example, formula, or code with explanation. 

### [Specific, Descriptive Rule Name]

[One sentence that states the rule clearly in plain language.]

[One to three sentences on when it applies, why it exists, and any important exceptions.]

[Optional: Example, formula, or code block — always preceded by a prose explanation.]

Common Mistakes to Avoid

• Orphan text: Text outside of defined chunks may not be retrieved correctly 
• Vague descriptions: Generic language reduces clarity and retrieval accuracy 
• Dependent content: Each chunk must stand alone without references to other sections 
• Code without explanation: Code alone does not provide enough context for accurate retrieval 

Pre-Publish Checklist

Agent Before saving your Knowledge Base:

• All section headings are clear and descriptive 
• Each section contains at least one chunk 
• No text exists outside structured headings 
• Each chunk covers a single concept 
• Every chunk begins with a clear summary 
• Key terms are repeated naturally within the content 
• No chunk depends on another to be understood 
• All examples and code are explained in plain language 

Outcome

A well-structured Knowledge Base improves how Denzing retrieves and interprets information. 

By organizing knowledge into clear, independent units, Agents can consistently surface the right context, leading to more accurate answers, better reasoning, and stronger alignment with your business.