Skip to main content

Keyword-Based Routing

Work in Progress

This use case describes a proposed feature that is not yet implemented in the current version of the semantic router. This represents a future capability that would enhance the router's functionality.

The Enterprise Challenge​

Imagine you're the CTO of a financial services company. Your team has built an AI-powered customer service system that routes queries to different AI models based on their content. However, you're facing critical challenges:

  • Security Risk: Customer queries about "internal account procedures" or "confidential trading strategies" are being sent to external AI providers, potentially exposing sensitive information
  • Compliance Violations: Regulatory requirements mandate that certain types of financial data must never leave your on-premise infrastructure
  • Lack of Control: You have no visibility into why certain routing decisions are made, making it impossible to audit or explain routing logic to regulators

The Solution: Keyword-Based Routing​

Keyword-based routing provides deterministic, transparent routing rules that complement AI-powered semantic classification. It allows enterprises to implement business policies with complete visibility and control.

Real-World Use Cases​

Use Case 1: Financial Services Data Sovereignty​

Scenario: A bank needs to ensure that any query containing internal procedures, account details, or trading strategies stays within their secure, on-premise infrastructure.

Business Rule: "Any query containing keywords like 'internal', 'confidential', 'account', 'trading', or 'procedures' must be routed to our internal AI model, never to external providers."

Implementation:

routing_rules:
  - name: "financial-confidential"
    description: "Route confidential financial queries to internal infrastructure"
    conditions:
      - type: "keyword_match"
        keywords: ["internal", "confidential", "account", "trading", "procedures", "strategy"]
    action:
      type: "route"
      endpoint: "internal-financial-ai"
      reasoning: "Contains confidential financial terminology - must stay on-premise"

Business Value:

  • ✅ Compliance with financial regulations
  • ✅ Complete data sovereignty
  • ✅ Audit trail for regulatory reporting
  • ✅ Zero risk of data leakage to external providers

Use Case 2: Healthcare Information Protection​

Scenario: A healthcare provider needs to route patient-related queries to HIPAA-compliant models while allowing general medical information queries to go to external providers with broader medical knowledge.

Business Rule: "Queries containing patient identifiers, medical records, or specific patient information go to HIPAA-compliant internal models. General medical questions can use external medical AI."

Implementation:

routing_rules:
  - name: "patient-data-protection"
    description: "Route patient-specific queries to HIPAA-compliant infrastructure"
    conditions:
      - type: "keyword_match"
        keywords: ["patient", "medical record", "diagnosis", "treatment plan", "prescription"]
      - type: "pii_detection"
        threshold: 0.7
    action:
      type: "route"
      endpoint: "hipaa-compliant-ai"
      reasoning: "Contains patient information - HIPAA compliance required"
  
  - name: "general-medical"
    description: "Route general medical questions to external medical AI"
    conditions:
      - type: "keyword_match"
        keywords: ["symptoms", "treatment", "medication", "disease", "condition"]
    action:
      type: "route"
      endpoint: "external-medical-ai"
      reasoning: "General medical query - can use external medical knowledge"

Use Case 3: Search Capability Routing​

Scenario: An enterprise wants to route queries that require web search to providers with search capabilities, while keeping other queries on their preferred models.

Business Rule: "Queries asking for 'search', 'find', 'look up', or 'current information' should go to providers with web search capabilities."

Implementation:

routing_rules:
  - name: "search-queries"
    description: "Route search requests to providers with web search capabilities"
    conditions:
      - type: "keyword_match"
        keywords: ["search", "find", "look up", "current", "latest", "recent", "news"]
    action:
      type: "route"
      endpoint: "search-enabled-provider"
      reasoning: "Query requires web search capabilities"

How It Works​

Keyword-based routing provides transparent, interpretable routing decisions by evaluating clear business rules before falling back to semantic classification:

  1. Rule Evaluation: Check each rule's conditions against the incoming query using simple keyword matching
  2. Match Found: Route to the specified endpoint with complete transparency - you know exactly why the decision was made
  3. No Match: Fall back to semantic classification for intelligent routing
  4. Audit Trail: Log every decision with clear reasoning for compliance and debugging

The key benefit is interpretability: stakeholders can easily understand and validate routing decisions because they're based on explicit, human-readable rules rather than opaque ML models.

Business Benefits​

Immediate Value​

  • Security: Ensure sensitive data never leaves your infrastructure with transparent rules
  • Compliance: Meet regulatory requirements with auditable, interpretable routing decisions
  • Capability Routing: Route queries to providers with specific capabilities (search, coding, etc.)
  • Performance: Get sub-millisecond routing for deterministic cases with clear reasoning

Enterprise Governance​

  • Transparency: See exactly why each query was routed where - no black box decisions
  • Interpretability: Human-readable rules that business stakeholders can understand and validate
  • Control: Implement business policies with confidence and clear justification
  • Auditability: Complete logs for compliance and regulatory reporting
  • Flexibility: Easy to modify rules as business needs change with full visibility

Getting Started​

Step 1: Identify Your Business Rules​

Start by identifying the key business policies that need deterministic routing:

  • Security Rules: What data must never leave your infrastructure?
  • Compliance Rules: What regulatory requirements do you need to meet?
  • Capability Rules: Which queries need specific capabilities (search, coding, etc.)?

Step 2: Define Your Keywords​

Create keyword lists for each business rule:

# Example: Financial services keywords
financial_keywords:
  confidential: ["internal", "confidential", "proprietary", "restricted"]
  account_data: ["account", "balance", "transaction", "statement"]
  trading: ["trading", "strategy", "portfolio", "investment"]
  procedures: ["procedure", "process", "workflow", "policy"]

Step 3: Configure Your Rules​

Define your routing rules with clear business justification:

routing_rules:
  - name: "confidential-financial"
    description: "Route confidential financial queries to internal infrastructure"
    business_justification: "Compliance with financial data sovereignty regulations"
    owner: "compliance-team@company.com"
    conditions:
      - type: "keyword_match"
        keywords: ["internal", "confidential", "account", "trading"]
    action:
      type: "route"
      endpoint: "internal-financial-ai"
      reasoning: "Contains confidential financial terminology"

Step 4: Test and Validate​

Test your rules with real queries to ensure they work as expected:

test_cases:
  - input: "How do I access internal account procedures?"
    expected_endpoint: "internal-financial-ai"
    expected_reasoning: "Contains 'internal' and 'account' keywords"
  
  - input: "What's the weather like today?"
    expected_endpoint: "semantic_classification"
    expected_reasoning: "No keyword matches - using semantic routing"

Success Stories​

Financial Services Company​

Challenge: Needed to ensure confidential trading strategies never reached external AI providers.

Solution: Implemented keyword-based routing for terms like "trading", "strategy", "confidential".

Result:

  • ✅ 100% compliance with data sovereignty requirements
  • ✅ Complete audit trail for regulatory reporting
  • ✅ Zero incidents of data leakage
  • ✅ 50% reduction in compliance review time

Healthcare Provider​

Challenge: Required HIPAA-compliant routing for patient data while using external AI for general medical knowledge.

Solution: Created rules to route patient-specific queries internally and general medical questions externally.

Result:

  • ✅ Full HIPAA compliance
  • ✅ Improved patient care with broader medical knowledge
  • ✅ Reduced costs by using appropriate models for each query type
  • ✅ Streamlined compliance audits

Next Steps​

  1. Evaluate Your Needs: Identify which business rules require deterministic routing
  2. Start Simple: Begin with high-priority security and compliance rules
  3. Iterate: Add more rules as you identify additional business requirements
  4. Monitor: Use audit trails to validate rule effectiveness and compliance
  5. Optimize: Refine rules based on real-world usage patterns