Certified Voice AI Architect

đź“ť Take Exam: Start Level 3 Written Exam
Accept the assignment to get your exam repository. Submit answers by creating an issue.

   
Duration 75 minutes
Questions 50
Passing Score 80% (40 correct)

Section A: Architecture Patterns (12 questions)

Question 1: Which pattern uses a central entry point to distribute calls to specialized agents?

  • A) Skill Composition
  • B) Gateway Router
  • C) Context State Machine
  • D) DataMap Integration

Question 2: When should you use the Skill Composition pattern?

  • A) When you need call routing logic
  • B) When you have common functionality across agents
  • C) When you need complex state transitions
  • D) When you need serverless API integration

Question 3: Which is an example of the “Monolithic Agent” anti-pattern?

  • A) Using environment variables for configuration
  • B) One agent with 50 functions covering all domains
  • C) Using multiple specialized agents
  • D) Implementing health checks

Question 4: What is the recommended approach for sharing state in horizontally scaled agents?

  • A) Store state in local memory
  • B) Use global variables
  • C) Externalize state to Redis or a database
  • D) Store state in the SWML document

Question 5: Which anti-pattern exposes implementation details to callers?

  • A) Hardcoded Configuration
  • B) Synchronous External Calls
  • C) Exposing Internal Errors
  • D) Monolithic Agent

Question 6: What is an Architecture Decision Record (ADR)?

  • A) A log of all API calls
  • B) A document explaining why a decision was made
  • C) A list of system errors
  • D) A performance benchmark

Question 7: Which principle suggests agents should have focused, specific responsibilities?

  • A) Defense in Depth
  • B) Single Responsibility
  • C) Fail Secure
  • D) Least Privilege

Question 8: What is the primary benefit of the Context State Machine pattern?

  • A) Better API integration
  • B) Structured multi-step workflows
  • C) Faster response times
  • D) Easier deployment

Question 9: Which scaling pattern requires stateless agents?

  • A) Vertical scaling
  • B) Horizontal scaling
  • C) Cold start scaling
  • D) Serverless scaling

Question 10: What should be documented in agent documentation?

  • A) Only the agent name
  • B) Endpoints, functions, dependencies, and configuration
  • C) Just the code
  • D) Marketing materials

Question 11: Which is NOT a recommended practice for agent architecture?

  • A) Use environment variables for secrets
  • B) Implement health checks
  • C) Put all business logic in one agent
  • D) Document architectural decisions

Question 12: What is the purpose of a Gateway Router in a multi-agent system?

  • A) Process payments
  • B) Route calls to appropriate specialized agents
  • C) Store customer data
  • D) Generate reports

Section B: Knowledge and RAG (8 questions)

Question 13: What does RAG stand for?

  • A) Rapid Agent Generation
  • B) Retrieval-Augmented Generation
  • C) Remote API Gateway
  • D) Real-time Audio Gateway

Question 14: Which CLI tool builds search indexes for knowledge bases?

  • A) swaig-test
  • B) sw-agent-init
  • C) sw-search
  • D) sw-index

Question 15: What is the purpose of “chunking” in knowledge base preparation?

  • A) Compress the data
  • B) Break content into searchable segments
  • C) Encrypt sensitive information
  • D) Validate document format

Question 16: Which parameter controls how many search results are returned?

  • A) distance
  • B) count
  • C) limit
  • D) max_results

Question 17: What is the distance parameter in search configuration?

  • A) Network latency
  • B) Similarity threshold for results
  • C) Maximum document size
  • D) Search timeout

Question 18: How can you combine static knowledge with live data?

  • A) You cannot combine them
  • B) Use search skill for static, DataMap for live
  • C) Put all data in the knowledge base
  • D) Only use live APIs

Question 19: What should you do when the knowledge base doesn’t have an answer?

  • A) Make up an answer
  • B) Admit limits and offer alternatives
  • C) Hang up the call
  • D) Repeat the question

Question 20: What is the recommended chunk size for FAQ content?

  • A) 50 tokens
  • B) 200 tokens
  • C) 1000 tokens
  • D) 5000 tokens

Section C: Performance Optimization (8 questions)

Question 21: What is a “cold start” in serverless deployments?

  • A) Running without cache
  • B) Initial container/runtime startup latency
  • C) Server reboot
  • D) Network timeout

Question 22: Which technique keeps users engaged during slow operations?

  • A) Hangup
  • B) Fillers
  • C) Transfer
  • D) Recording

Question 23: What is the recommended timeout for external API calls in voice agents?

  • A) 30 seconds
  • B) 2-5 seconds
  • C) 1 minute
  • D) No timeout needed

Question 24: Which caching decorator is built into Python?

  • A) @cache
  • B) @lru_cache
  • C) @memoize
  • D) @cached

Question 25: How can you run multiple warehouse checks in parallel?

  • A) Sequential loops
  • B) ThreadPoolExecutor
  • C) Longer timeouts
  • D) Multiple agents

Question 26: What makes prompts “optimized” for performance?

  • A) Using more words
  • B) Being concise and focused
  • C) Including all possible scenarios
  • D) Using complex language

Question 27: What is the purpose of the end_of_speech_timeout parameter?

  • A) Call duration limit
  • B) Time to wait after user stops speaking
  • C) Function timeout
  • D) Recording duration

Question 28: Which metric type is best for measuring function latency?

  • A) Counter
  • B) Gauge
  • C) Histogram
  • D) Summary

Section D: Security and Compliance (10 questions)

Question 29: What should happen to call recording when collecting credit card numbers?

  • A) Continue recording
  • B) Pause recording
  • C) Stop recording permanently
  • D) Increase volume

Question 30: What is the maximum number of PIN attempts before lockout (best practice)?

  • A) 1
  • B) 3
  • C) 10
  • D) Unlimited

Question 31: Which parameter marks a function as handling sensitive data?

  • A) private=True
  • B) secure=True
  • C) sensitive=True
  • D) protected=True

Question 32: What should be stored in logs regarding credit card numbers?

  • A) Full card number
  • B) Encrypted card number
  • C) Last 4 digits only
  • D) Nothing about cards

Question 33: Which type of authentication protects SWML endpoints?

  • A) OAuth
  • B) Basic Auth
  • C) API Keys only
  • D) No authentication needed

Question 34: What is tokenization in payment processing?

  • A) Breaking text into words
  • B) Replacing card numbers with secure tokens
  • C) Encrypting all data
  • D) Validating card format

Question 35: Which input pattern might indicate prompt injection?

  • A) “What time is it?”
  • B) “ignore previous instructions”
  • C) “What’s my balance?”
  • D) “Transfer to sales”

Question 36: What should happen when an account is locked due to failed PIN attempts?

  • A) Keep trying
  • B) End the call with guidance
  • C) Transfer to sales
  • D) Reset the PIN automatically

Question 37: What is the purpose of timing-safe comparison for PIN verification?

  • A) Faster verification
  • B) Prevent timing attacks
  • C) Better logging
  • D) Simpler code

Question 38: Which data should NEVER be in logs?

  • A) Call ID
  • B) Timestamps
  • C) Full credit card numbers
  • D) Function names

Section E: Monitoring and Troubleshooting (12 questions)

Question 39: What are the three pillars of observability?

  • A) Speed, Size, Security
  • B) Logs, Metrics, Traces
  • C) CPU, Memory, Network
  • D) Input, Output, Errors

Question 40: Which log format enables easy querying and analysis?

  • A) Plain text
  • B) Structured JSON
  • C) CSV
  • D) XML

Question 41: What is a correlation ID used for?

  • A) Measuring latency
  • B) Tracking requests across systems
  • C) Counting errors
  • D) Storing state

Question 42: Which Prometheus metric type tracks current values (like active calls)?

  • A) Counter
  • B) Gauge
  • C) Histogram
  • D) Summary

Question 43: What should a health check endpoint verify?

  • A) Only that the server responds
  • B) Agent, database, and external API status
  • C) Just CPU usage
  • D) Only memory

Question 44: Which alert would indicate potential AI handling issues?

  • A) High CPU usage
  • B) High transfer rate
  • C) Low memory
  • D) High disk usage

Question 45: What does histogram_quantile(0.95, ...) calculate?

  • A) Average latency
  • B) 95th percentile latency
  • C) Maximum latency
  • D) Minimum latency

Question 46: What is the purpose of the swaig-test command --dump-swml?

  • A) Test a function
  • B) Output the SWML document
  • C) List all tools
  • D) Start the server

Question 47: How do you test a specific function with swaig-test?

  • A) --exec function_name
  • B) --test function_name
  • C) --call function_name
  • D) --run function_name

Question 48: What indicates a “degraded” health status?

  • A) All checks pass
  • B) Some checks fail but system works
  • C) Complete system failure
  • D) Network timeout

Question 49: Which pattern helps handle temporary external service failures?

  • A) Immediate failure
  • B) Circuit breaker
  • C) Ignore errors
  • D) Infinite retry

Question 50: What should logs EXCLUDE for security?

  • A) Timestamps
  • B) Sensitive data (passwords, card numbers)
  • C) Function names
  • D) Call IDs

Submission

Submit your answers via GitHub Classroom. Your exam will be auto-graded.

Scoring

  • 45-50 correct: Excellent
  • 40-44 correct: Good (Pass)
  • Below 40: Review required

Next: Level 3 Practical Exam

Back to top

SignalWire AI Agents Certification Program