Compare Azure services and design choices

Key differences

• A project is a workspace-level organizing concept; a service resource is the deployable Azure control-plane boundary.
• Resources carry keys, endpoints, managed identity, private endpoints, and diagnostic settings.
• Projects help learners reason about app lifecycle, but production security still lives at the resource, identity, and network layers.

Common mistake

Assuming a project setting secures all deployed endpoints. Exam scenarios often separate development workspace choices from resource-level access and network controls.

Check before deciding

1. Identify whether the question is asking about development organization or runtime endpoint control.
2. Check service resource keys, identity, network rules, and deployments before release.
3. Open the AI services runbook before changing keys, networking, or model deployments.

Related terms

Key differences

• Indexers depend on data-source credentials, skillsets, schedules, and indexer status.
• Manual ingestion shifts transformation, retries, and error handling into app/pipeline code.
• Exam questions often reveal the answer through data-source support, refresh cadence, or required transformations.

Common mistake

Debugging query relevance while ignoring indexer failures or stale source-to-index refresh status.

Check before deciding

1. Check indexer status and error messages before changing index schema.
2. Confirm whether the source is supported by an indexer.
3. Use manual ingestion when transformations or event timing require app ownership.

Related terms

Key differences

• Keys are simple but can leak and must be rotated.
• Managed identity reduces secret handling but requires role assignment and supported service integration.
• Exam scenarios often mention Key Vault, app settings, identity, or least privilege as clues.

Common mistake

Putting an AI service key directly in app settings or code because it made the first demo work.

Check before deciding

1. Check whether the target AI service supports identity-based access for the operation.
2. If keys are used, store them in Key Vault and rotate them.
3. List role assignments and app identity before changing access.

Related terms

Key differences

• Content Safety focuses on content category risk; Prompt Shields focus on attack patterns against instructions and grounding.
• Production systems may need both, plus human review and logging.
• Prompt-injection risk becomes more important when agents, tools, or retrieval are involved.

Common mistake

Assuming content moderation alone protects agent tools or system prompts from malicious retrieved text.

Check before deciding

1. Classify the threat: harmful content, prompt injection, data leakage, or unsafe tool action.
2. Log and review blocked/flagged cases before loosening controls.
3. Test with adversarial prompts and retrieved content, not only normal happy-path prompts.

Related terms

Key differences

• AI Search owns indexes, analyzers, vector fields, data sources, indexers, and retrieval ranking.
• Azure OpenAI owns model deployments, prompt behavior, token usage, and generation quality.
• Many production designs use both: Search retrieves evidence, then OpenAI produces the answer grounded on that evidence.

Common mistake

Treating Azure OpenAI as a knowledge base. Without retrieval, permissions, and grounding, model output may not reflect your approved data.

Check before deciding

1. List the data sources, indexes, and model deployments involved before choosing the pattern.
2. Test a real query with expected citations or source IDs, not only a demo prompt.
3. Check indexing cost, token cost, access controls, and private-network requirements before production rollout.

Related terms

Key differences

• Vision OCR is image-centric; Document Intelligence is document-centric.
• Document Intelligence adds layout, prebuilt models, and custom extraction choices.
• The right answer often depends on whether the exam scenario asks for fields/tables or raw text.

Common mistake

Using generic OCR for document workflows that require field confidence, tables, or model-specific extraction.

Check before deciding

1. Identify whether the input is an image, document, form, or invoice.
2. Check whether raw text is enough or structured fields are required.
3. Review endpoint, key, region, and network access before deployment.

Related terms

Key differences

• CLU is dialog/intent oriented; classification is category-label oriented.
• CLU design depends on utterances, intents, and entities.
• Classification design depends on labeled examples and category boundaries.

Common mistake

Using a conversational intent model for every NLP problem even when the requirement is document classification.

Check before deciding

1. Read the requirement for intents, entities, categories, and expected output.
2. Choose sample data that mirrors real user utterances or documents.
3. Check endpoint, deployment, and confidence thresholds before production.

Related terms

Key differences

• Layout is broader and structure-oriented; prebuilt models are targeted and field-oriented.
• Prebuilt models reduce custom training when the document type matches.
• Layout is often a step before deciding whether custom extraction is needed.

Common mistake

Training a custom model before testing whether layout or a prebuilt model already solves the extraction need.

Check before deciding

1. Identify the document type and fields required.
2. Try layout/prebuilt extraction with sample documents.
3. Use custom extraction only when the generic or prebuilt path misses required fields.

Related terms

Key differences

• Agents add orchestration concepts that must be secured and monitored beyond the model call.
• Chat completion remains simpler and easier to reason about for stateless prompt/response flows.
• Agent scenarios usually require more attention to tool permissions, data boundaries, and human approval points.

Common mistake

Using an agent for every generative AI scenario. On exams, the simpler model call is often enough when no tool orchestration is required.

Check before deciding

1. List tools, data sources, and actions the agent can call.
2. Separate model output risk from tool-execution risk.
3. Check identity and approval boundaries before allowing an agent to change data or call APIs.

Related terms

Key differences

• Integrated vectorization reduces custom code but still needs service identity, indexer, skillset, and data-source permissions.
• Custom pipelines can fit mature data engineering flows but create more moving parts to secure and monitor.
• Both patterns need evaluation for chunk quality, relevance, cost, and stale index behavior.

Common mistake

Choosing vectorization mechanics before defining retrieval quality, permissions, refresh cadence, and cost limits.

Check before deciding

1. Check data source, index, vector fields, and skillset permissions.
2. Measure retrieval quality with expected queries and citations.
3. Review indexer failures and service scale before production.

Related terms

Key differences

• Catalog selection is design-time; deployments are runtime resources with cost, quotas, and access patterns.
• Applications usually need deployment names/endpoints rather than catalog entries.
• Operational questions focus on metrics, throttling, capacity, and deployment health.

Common mistake

Assuming selecting a model in a catalog means the application is ready to call it in production.

Check before deciding

1. Identify model, deployment name, region, quota, and endpoint.
2. Check tokens, throttling, metrics, and content safety before release.
3. Keep official model availability and pricing checks outside the static exam notes.

Related terms

Key differences

• Online endpoints need traffic routing, authentication, scale, and live availability planning.
• Batch endpoints need job scheduling, input/output storage, retry behavior, and throughput planning.
• Online cost is tied to serving capacity; batch cost is tied to job execution and data movement.

Common mistake

Using an online endpoint for large offline jobs, then fighting timeout, cost, and scaling issues that batch inference is designed to absorb.

Check before deciding

1. Confirm latency, throughput, authentication, and failure-retry expectations with the consuming app owner.
2. Inspect existing endpoint deployments and traffic before changing production inference routes.
3. Run a representative test payload and record latency, error rate, and cost before committing.

Related terms

Key differences

• Prebuilt models reduce training effort but limit field control.
• Custom models require labeled examples and validation but can fit your document shape.
• Custom accuracy depends on representative samples, versioning, and ongoing review after document templates change.

Common mistake

Training a custom model before proving that a prebuilt model fails on real samples, or trusting a prebuilt model without testing edge cases.

Check before deciding

1. Run a sample set through the prebuilt model and record fields, confidence scores, and misses.
2. If custom is needed, separate training, validation, and holdout samples before production use.
3. Check data privacy, region, and cost before sending production documents through any model.

Related terms

Key differences

• Private endpoint answers “where can traffic come from?”
• Managed identity answers “who is allowed to call or manage this?”
• Strong designs often need both, plus DNS, diagnostics, and least-privilege roles.

Common mistake

Adding a private endpoint and assuming identity/role assignments are no longer needed.

Check before deciding

1. Separate network controls from identity controls in the scenario.
2. Validate private DNS resolution before disabling public access.
3. Review role assignments and keys before release.

Related terms

Key differences

• Prompt flow helps with experimentation and evaluation; code owns production-grade control and app-specific lifecycle.
• Code-first paths may integrate better with existing CI/CD and testing standards.
• Prompt flow can clarify model and prompt behavior before teams harden the production implementation.

Common mistake

Treating prompt flow as a complete production architecture without checking identity, observability, deployment, and rollback requirements.

Check before deciding

1. Map the flow steps to the app runtime path.
2. Check how secrets, identities, and telemetry are handled.
3. Use a deployment runbook before promoting prompt or code changes.

Related terms

Key differences

• RAG keeps knowledge in data sources and indexes; fine-tuning changes the model artifact or deployment behavior.
• RAG needs retrieval quality, chunking, indexing, and access control.
• Fine-tuning needs training data quality, evaluation, cost, deployment governance, and rollback planning.

Common mistake

Fine-tuning a model just because the model does not know private company documents. That is usually a retrieval/grounding problem first.

Check before deciding

1. Identify whether the exam scenario asks for current facts, citations, or private data.
2. Test retrieval precision and permission filtering before changing model behavior.
3. Compare token, index, and training costs before production.

Related terms

Key differences

• An indexer controls data source reads, schedules, field mappings, and indexing runs.
• A skillset controls enrichment steps that transform or add fields during indexing.
• A failed indexer run may be caused by data access, mapping, or skill execution, so inspect both when troubleshooting.

Common mistake

Adding a skillset to fix a data-loading problem. If the indexer cannot reach or map the source, enrichment will not solve it.

Check before deciding

1. List the data source, index, indexer, and skillset before editing any one of them.
2. Run or inspect indexer status in a non-production window if reindexing can change visible search results.
3. Capture failing field names, skill errors, and document counts before changing mappings.

Related terms

Key differences

• Speech to text is the base transcription workload.
• Speech translation adds language translation requirements and related quality considerations.
• Both require attention to language, audio quality, region, latency, and privacy requirements.

Common mistake

Choosing translation when the scenario only asks for transcription or captions in the original language.

Check before deciding

1. Identify source language, target language, and latency requirement.
2. Test representative audio, accents, and noise conditions.
3. Check endpoint and key handling before release.

Related terms

Key differences

• Vector search depends on embeddings, vector fields, chunking strategy, and distance similarity.
• Semantic ranker works over text results and semantic configuration; it does not replace an embedding strategy.
• Hybrid retrieval can combine keyword, vector, and semantic ranking, but each layer adds cost and tuning complexity.

Common mistake

Turning on every relevance feature at once and then not knowing whether vector quality, keyword matching, or semantic ranking caused the result.

Check before deciding

1. Create a small relevance test set with expected documents before changing ranking behavior.
2. Check index fields, vector dimensions, semantic configuration, and query mode before production changes.
3. Measure quality and latency separately for keyword, vector, semantic, and hybrid queries.

Related terms