Practice Test Geeks(CAIC) Certified AI Consultant Practice Test
← All CAIC Flashcard Decks

Cloud Infrastructure Flashcards

7 cards from real CAIC practice questions. Tap to flip, then mark Knew It or Still Learning — missed cards come back until you master them.

Read the first 7 Cloud Infrastructure flashcards as text

  1. Which cloud observability pillar provides time-series numeric measurements used to trigger auto-scaling for AI inference services?

    Answer: Metrics

    Metrics (e.g., CPU utilization, GPU memory usage, request latency p99) are the numeric signals consumed by auto-scalers and alerting systems.

  2. A CAIC consultant is evaluating cloud AI services. What is the key trade-off when choosing a fully managed AI platform (e.g., SageMaker, Vertex AI) over a self-built solution?

    Answer: Managed platforms reduce operational burden but introduce vendor lock-in and less customization

    Managed platforms abstract infrastructure complexity and accelerate time-to-market but tie you to provider APIs, making migration to alternatives costly.

  3. Which storage technology is most appropriate for storing high-dimensional vector embeddings used in AI similarity search at scale?

    Answer: Vector database (e.g., Pinecone, Weaviate, pgvector)

    Vector databases are optimized for approximate nearest-neighbor (ANN) search over high-dimensional embeddings, enabling fast semantic similarity queries.

  4. What cloud networking feature allows an AI workload on a private subnet to download model packages from the internet WITHOUT exposing the instances to inbound internet traffic?

    Answer: NAT Gateway

    A NAT Gateway allows outbound internet access from private subnets while blocking unsolicited inbound connections, maintaining security posture.

  5. When should an AI consultant recommend using serverless inference (e.g., AWS Lambda, Google Cloud Run) instead of a dedicated GPU instance?

    Answer: For lightweight, infrequent inference tasks where cold starts are acceptable

    Serverless is cost-effective for sporadic, lightweight inference (e.g., small classifiers) where paying for idle GPU capacity would be wasteful.

  6. Which cloud design pattern uses a read-optimized replica of a datastore to serve AI feature queries without impacting the transactional write path?

    Answer: CQRS (Command Query Responsibility Segregation)

    CQRS separates read and write models, allowing a dedicated read replica optimized for feature serving queries while writes go to a normalized transactional store.

  7. A company wants to ensure their cloud-hosted AI models comply with GDPR data residency requirements. Which infrastructure control directly addresses this?

    Answer: Pinning workloads and data storage to specific geographic regions using cloud region selectors and data boundary policies

    Cloud region selectors and data boundary configurations (e.g., AWS Data Perimeter, Azure Geo-restriction policies) ensure data never leaves approved jurisdictions.