RATIO MACHINA STARTER logo

MERCURY

Knowledge Base Configuration

Knowledge bases are the foundation of intelligent conversational AI. By providing your personas with comprehensive, structured knowledge through document uploads and URL fetching, you enable them to deliver accurate, contextually relevant responses across any domain or use case.

What is a Knowledge Base?

A knowledge base in Mercury is a structured repository of information that your personas can access to provide accurate, relevant responses. Think of it as giving your AI agents a comprehensive library of information they can reference during conversations.

🧠 Key Concept: Unlike simple keyword matching, Mercury's knowledge bases use advanced semantic understanding to find relevant information even when questions don't match exact terms in your documents.

How Knowledge Bases Work

  1. Content Ingestion: Documents and web pages are processed and indexed
  2. Semantic Vectorization: Content is converted into searchable vector representations
  3. Query Processing: When users ask questions, similar vectors are identified
  4. Context Integration: Relevant information is seamlessly integrated into responses

Benefits for Your Personas

  • Accuracy: Responses based on your authoritative content
  • Consistency: Same information delivered across all conversations
  • Scalability: Handle complex queries without human intervention
  • Currency: Always up-to-date with your latest information

Types of Knowledge Sources

Mercury supports multiple knowledge source types, each optimized for different content formats and use cases:

Document-Based Knowledge

Supported Formats

  • PDF Documents: Manuals, guides, reports
  • Word Documents: Policies, procedures, FAQs
  • Markdown Files: Technical documentation
  • Text Files: Knowledge articles, scripts
  • Spreadsheets: Product catalogs, pricing
  • PowerPoint: Training materials, presentations

Best Use Cases

  • Product documentation and manuals
  • Company policies and procedures
  • Training materials and guidelines
  • FAQs and knowledge articles
  • Legal documents and compliance guides
  • Technical specifications

Processing Features

  • Automatic text extraction from images
  • Table and structure preservation
  • Metadata extraction (title, author, date)
  • Section and heading recognition
  • Multi-language support
  • Version tracking and updates

Web-Based Knowledge

Content Sources

  • Company Websites: Product pages, about sections
  • Knowledge Bases: Help centers, wikis
  • Blog Content: Articles, guides, tutorials
  • API Documentation: Developer resources
  • Forums: Community Q&A content
  • News Sites: Industry updates, announcements

Fetching Methods

  • Single URL ingestion
  • Sitemap-based crawling
  • Recursive page following
  • RSS feed monitoring
  • Scheduled refresh updates
  • Change detection and incremental updates

Content Processing

  • Clean HTML extraction
  • Navigation and footer removal
  • Article content identification
  • Image alt-text extraction
  • Link preservation and context
  • Duplicate content detection

Structured Data Sources

Database Integration

  • Customer information databases
  • Product catalogs and inventory
  • Order history and transaction data
  • Support ticket databases
  • User profiles and preferences

API Data Sources

  • CRM system data (Salesforce, HubSpot)
  • E-commerce platforms (Shopify, WooCommerce)
  • Support systems (Zendesk, Intercom)
  • Analytics platforms (Google Analytics)
  • Business intelligence tools

Planning Your Knowledge Strategy

Before implementing your knowledge base, it's important to plan your approach based on your personas' roles and the types of questions they'll handle.

Content Audit & Inventory

Identify Existing Resources

Content Assessment Checklist
  • □ Company knowledge base articles
  • □ Product documentation and manuals
  • □ FAQ sections and help content
  • □ Training materials and guides
  • □ Policy documents and procedures
  • □ Marketing materials and case studies
  • □ Blog posts and educational content
  • □ Technical specifications and API docs
  • □ Support ticket solutions and resolutions

Quality Assessment

Content Quality Criteria
  • □ Accuracy: Information is current and correct
  • □ Completeness: Topics are thoroughly covered
  • □ Clarity: Content is well-written and understandable
  • □ Relevance: Information matches persona's role
  • □ Authority: Content comes from reliable sources
  • □ Freshness: Information is up-to-date
  • □ Structure: Content is well-organized
  • □ Searchability: Key terms are present

Persona-Specific Knowledge Planning

Customer Support Persona

Essential Knowledge
  • Product troubleshooting guides
  • Common issue resolutions
  • Account management procedures
  • Escalation policies and workflows
  • Billing and payment information
  • Service level agreements
Knowledge Sources
  • Support knowledge base articles
  • Product documentation
  • Internal procedure documents
  • FAQ sections
  • Training materials

Sales Persona

Essential Knowledge
  • Product features and benefits
  • Pricing and packaging options
  • Competitive comparisons
  • Case studies and success stories
  • Objection handling scripts
  • Qualification criteria
Knowledge Sources
  • Sales playbooks and materials
  • Product marketing content
  • Competitive analysis documents
  • Customer testimonials
  • Pricing sheets and proposals

Technical Persona

Essential Knowledge
  • API documentation and examples
  • Integration guides and tutorials
  • Code samples and libraries
  • Architecture diagrams
  • Security and compliance guides
  • Troubleshooting procedures
Knowledge Sources
  • Technical documentation sites
  • Developer portals
  • Code repositories (README files)
  • Technical blog posts
  • Architecture documentation

Optimization & Best Practices

Knowledge Quality Management

Content Curation Guidelines

✅ Do Include
  • Authoritative, official documentation
  • Up-to-date product information
  • Well-structured, comprehensive guides
  • Frequently asked questions and answers
  • Step-by-step procedures and tutorials
  • Troubleshooting and solution articles
❌ Avoid Including
  • Outdated or deprecated information
  • Personal opinions or speculative content
  • Duplicate or redundant information
  • Marketing fluff without substance
  • Internal-only or confidential information
  • Unstructured or poorly written content

Quality Metrics & Monitoring

{
  "qualityMetrics": {
    "contentFreshness": {
      "maxAge": "90d",
      "warningThreshold": "30d",
      "alertThreshold": "60d"
    },
    "relevanceScoring": {
      "minScore": 0.7,
      "evaluationFrequency": "weekly",
      "improvementActions": ["update", "remove"]
    },
    "usageAnalytics": {
      "trackQueries": true,
      "identifyGaps": true,
      "measureSatisfaction": true
    },
    "performanceMetrics": {
      "responseTime": "< 500ms",
      "accuracyRate": "> 95%",
      "coverageRate": "> 90%"
    }
  }
}

Performance Optimization

Indexing Optimization

  • Chunking Strategy: Optimize chunk size for your content type
  • Overlap Configuration: Balance context preservation with storage
  • Vector Dimensions: Choose appropriate embedding dimensions
  • Index Partitioning: Separate content by type or recency
  • Compression: Use efficient storage for large knowledge bases

Query Optimization

  • Semantic Search: Prioritize meaning over keyword matching
  • Query Expansion: Include synonyms and related terms
  • Context Filtering: Use conversation context to narrow results
  • Result Ranking: Weight results by relevance and recency
  • Caching Strategy: Cache frequent queries and results

Maintenance Practices

  • Regular Audits: Review and clean up outdated content
  • Performance Monitoring: Track query performance and accuracy
  • User Feedback: Collect and act on user satisfaction data
  • Content Lifecycle: Implement versioning and archiving
  • Security Updates: Ensure access controls and data protection

Security & Compliance

Data Protection

{
  "security": {
    "encryption": {
      "atRest": "AES-256",
      "inTransit": "TLS-1.3",
      "keyManagement": "customer_managed"
    },
    "accessControl": {
      "authentication": "required",
      "authorization": "role_based",
      "auditLogging": "enabled"
    },
    "dataClassification": {
      "sensitiveData": "redaction_enabled",
      "piiDetection": "automatic",
      "complianceLabels": ["GDPR", "CCPA"]
    },
    "retention": {
      "defaultPeriod": "7y",
      "autoArchival": "enabled",
      "rightToErasure": "supported"
    }
  }
}

Content Governance

  • Content Approval: Review process for sensitive content
  • Source Verification: Validate authenticity of information sources
  • Access Permissions: Control who can modify knowledge bases
  • Audit Trails: Track all changes and access patterns
  • Compliance Scanning: Automated detection of policy violations
  • Data Lineage: Track origin and modifications of content

Ready to Build Your Knowledge Base?

With comprehensive knowledge base configuration, your Mercury personas can provide accurate, contextual responses that truly serve your users' needs. Start with high-quality, curated content and expand using automated web fetching for dynamic information.

Next Steps

  1. Audit your existing content and identify knowledge gaps
  2. Start with document uploads for core, stable information
  3. Configure URL fetching for dynamic, web-based content
  4. Set up monitoring and update schedules
  5. Test and refine based on persona performance

Key Success Factors

  • Focus on quality over quantity of content
  • Keep information current and relevant
  • Structure content for optimal AI understanding
  • Monitor usage and performance metrics
  • Continuously improve based on user feedback
Continue to Quickstart →Explore APIs →
Transcend the hype with MERCURY by Ratio Machina