AI Bot
AI-500: Architecture & Design Patterns Flashcards
Microsoft Multi-Agent AI Solutions Expert AI-500 Flashcards
Study our AI-500: Architecture & Design Patterns flashcards for the Microsoft Multi-Agent AI Solutions Expert AI-500 exam with 31+ flashcards. View as flashcards, a searchable table, or as a fun matching game.
| Front | Back |
| Actor model vs agent model difference | Actor model focuses on message passing encapsulated state Agent model adds goals planning and reasoning |
| Agent discovery and registry pattern | Use a service registry or metadata store for available agents and capabilities |
| Azure Functions use case | Serverless event driven compute for small stateless processing tasks |
| Azure OpenAI typical responsibilities | Text generation summarization embeddings and few shot prompting |
| Blackboard pattern summary | Shared knowledge base where agents read write and coordinate via common state |
| Broker pattern in agent systems | Broker mediates communication decouples clients and agents |
| Cognitive Search and vector search mapping | Use Cognitive Search with semantic ranking and vector store for retrieval augmented generation |
| Cosmos DB when to use | Use Cosmos DB for low latency globally distributed application state |
| Cost vs performance optimization ideas | Use model selection batching caching and autoscaling to balance cost and QoS |
| Designing for extensibility tips | Keep clear component boundaries version APIs and use feature toggles for rollout |
| Durable Functions role in orchestration | Durable Functions implement reliable long running orchestrations with state management |
| Embedding storage strategies | Store embeddings in vector DB like Cognitive Search or Cosmos DB with index |
| Event driven topology benefit | Loose coupling asynchronous scaling and resilience to partial failures |
| Event Hubs typical use case | Ingest telemetry and streaming data at high scale |
| Failure handling strategies | Retries exponential backoff dead letter queues and compensating actions |
| Fan out fan in pattern use case | Parallel tasks executed then aggregated for faster throughput |
| Latency vs consistency trade off | Stronger consistency can increase latency choose per user experience requirements |
| Mediator pattern role | Mediator centralizes complex interaction logic between agents |
| Monitoring and observability for agents | Collect telemetry traces logs and metrics with Azure Monitor and Application Insights |
| Multi agent architecture pattern definition | System of autonomous agents collaborating through messages to achieve goals |
| Orchestration vs autonomy key difference | Orchestration uses a central coordinator Autonomy lets agents act independently |
| Pipeline pattern for LLM workflows | Sequential components each performing a transformation on the request |
| Pub sub pattern on Azure | Use Event Grid or Service Bus for event distribution and decoupled consumers |
| Reliable messaging choice Service Bus vs Event Grid | Use Service Bus for ordered durable messages Event Grid for high throughput event routing |
| Security best practices on Azure for AI | Use Key Vault managed identities role based access and network isolation |
| State management options on Azure | Use Cosmos DB Durable Entities or Blob Storage depending on consistency and latency needs |
| Stateless vs stateful components trade off | Stateless scales easily but needs external stateful stores for context |
| Testing multi agent systems approach | Use integration tests contract tests and scenario based simulations |
| Throughput vs accuracy trade off | Higher throughput may require model quantization smaller models at cost of accuracy |
| When to choose autonomy | Use autonomy for resilience local decision making and decentralized scaling |
| When to choose orchestration | Use orchestration for strict workflows transactions and centralized control |