AI Bot
AI-103: Azure OpenAI and Cognitive Services Flashcards
Microsoft Azure AI App and Agent Developer Associate AI-103 Flashcards
Study our AI-103: Azure OpenAI and Cognitive Services flashcards for the Microsoft Azure AI App and Agent Developer Associate AI-103 exam with 49+ flashcards. View as flashcards, a searchable table, or as a fun matching game.
| Front | Back |
| Best practice for metadata with embeddings? | Store metadata for filtering provenance and chunk ids |
| Common uses of embeddings? | Semantic search clustering and retrieval augmented generation |
| Difference between Azure OpenAI and Cognitive Services? | Azure OpenAI exposes LLMs while Cognitive Services offers specialized prebuilt models for vision speech and language |
| How to authenticate to Azure OpenAI? | Use Azure AD or API keys stored in Azure Key Vault |
| How to choose an embedding model? | Balance dimensionality accuracy and cost based on retrieval needs |
| How to chunk documents for embeddings? | Split by semantic boundaries and keep chunks under token limits |
| How to handle long documents with LLMs? | Use chunking embedding retrieval and RAG or summarization |
| How to handle PII in user inputs? | Mask or redact sensitive fields log minimal data and apply retention policies |
| How to integrate Azure OpenAI with Azure Functions? | Use managed identity or Key Vault in function to call APIs securely |
| How to manage rate limits and throttling? | Implement exponential backoff retries and request batching |
| How to monitor model performance in Azure? | Use Azure Monitor Application Insights and custom telemetry |
| How to reduce hallucinations? | Ground responses with retrieved context use verification and source attribution |
| How to version prompts and models? | Store prompt templates and deployment ids in source control and configuration |
| Mitigation for prompt injection? | Sanitize inputs apply system messages and validate retrieved documents |
| What are best practices for temperature for deterministic tasks? | Set temperature near zero for consistent outputs |
| What are Cognitive Services APIs? | Prebuilt cloud APIs for Vision Speech and Language tasks |
| What are deployment region considerations? | Model availability latency and compliance vary by Azure region |
| What are system messages in chat completions? | Instructions that set model behavior and role |
| What is a private endpoint? | Connect services over private network eliminating public internet exposure |
| What is a prompt in LLM usage? | Text input instructing the model what to do |
| What is an embedding? | Numeric vector representation of text for semantic similarity tasks |
| What is anomaly detection used for? | Detect deviations in metrics logs or sensor data |
| What is Azure Cognitive Search vector search? | Managed search service supporting vector and keyword queries |
| What is Azure Key Vault used for? | Securely store API keys secrets and certificates |
| What is Azure OpenAI Service? | Managed hosting of OpenAI models on Azure with enterprise security and compliance |
| What is cost optimization for LLM apps? | Use smaller models caching batching and hybrid retrieval to reduce spend |
| What is model deployment in Azure OpenAI? | Assign a model to a named deployment endpoint for API calls |
| What is OCR in Vision APIs? | Optical Character Recognition to extract text from images and documents |
| What is prompt chaining? | Link multiple prompts or model calls to achieve complex workflows |
| What is prompt injection risk? | Malicious input that tries to override instructions or leak data |
| What is RAG retrieval augmented generation? | Combine vector retrieval with LLM responses to ground answers in documents |
| What is semantic captioning in vision? | Create concise descriptions of images that capture meaning not just objects |
| What is semantic search vs lexical search? | Semantic uses meaning via embeddings lexical uses exact term matching |
| What is speaker diarization? | Identify who spoke when in multi speaker audio |
| What is speech to text used for? | Transcribe audio for captions search and voice interfaces |
| What is streaming for speech and text? | Continuous transfer of audio or tokens for real time applications |
| What is temperature parameter? | Controls randomness higher means more creative lower means more deterministic |
| What is text to speech used for? | Generate natural audio for assistants accessibility and notifications |
| What is the Chat Completions endpoint used for? | Conversational interactions using role based messages |
| What is the content moderation API used for? | Detect and filter harmful or policy violating content |
| What is the purpose of system prompts for safety? | Constrain model behavior to enforce safety and policy rules |
| What is top_p parameter? | Controls nucleus sampling probability mass for output diversity |
| What is zero shot vs few shot prompting? | Zero shot gives task only few shot provides examples in prompt |
| When to use fine tuning vs prompt engineering? | Fine tune for consistent domain behavior prompt engineering for rapid iteration |
| When to use streaming responses? | When you need partial tokens for low latency real time output |
| Why batch requests for embeddings? | Reduce latency and lower costs by processing multiple items per call |
| Why track tokens and costs? | Tokens determine billing and help optimize prompt length and cost |
| Why use cosine similarity? | Measure angular similarity between embedding vectors |
| Why use Managed Identity with Cognitive Services? | Enable secure keyless access from Azure resources |
About the Flashcards
Prepare effectively with these comprehensive Flashcards for the Microsoft Azure AI App and Agent Developer Associate exam. This deck is designed to help you master essential terminology, concepts, and key ideas related to cloud-based language and vision services. Students can review core topics such as prompt engineering, large language model parameters, and retrieval augmented generation.
By studying these flashcards, you will reinforce your understanding of text embeddings, semantic search, and operational best practices like cost optimization and handling rate limits. The deck also covers critical security configurations, including managed identities and private endpoints, ensuring you are fully prepared to tackle the assessment.
Topics covered in this flashcard deck:
- Azure OpenAI and Cognitive Services
- Prompt engineering and model parameters
- Embeddings and retrieval augmented generation
- Security and managed identities
- API performance and cost optimization