Pass Microsoft AI-103 Exam in First Attempt Easily
Real Microsoft AI-103 Exam Questions, Accurate & Verified Answers As Experienced in the Actual Test!

Verified by experts

AI-103 Premium File

  • 187 Questions & Answers
  • Last Update: Nov 19, 2026
$69.99 $76.99

Developing AI Apps and Agents on Azure AI-103 Practice Test 7

Question 60

A team implementing Microsoft Foundry faces this scenario: A product manager needs to attribute weekly agent spending to business units using separate deployments. The nonnegotiable requirement is to measure inference cost by workload rather than averaging all project charges. Which design choice best meets it?

  1. Disable token telemetry to reduce logging storage cost.
  2. Count the number of agent names and estimate identical costs for each.
  3. Tag deployments and correlate per-request model/token telemetry with workload identity and billing data.
  4. Replace usage logging with occasional screenshot sampling.

Explanation

Correct Answer

C:

Cost analysis needs attributable usage and model pricing, not a single total; for this workload the decision is to tag deployments and correlate per-request model/token telemetry with workload identity and billing data. The acceptance result is: Reconcile sample requests against measured prompt/output tokens and project cost records. An independent edge-case check is: Check that tool and retry costs are counted without double charging failed calls. That distinction matters because cross-service tool charges and cached-input discounts can make simplistic per-token estimates inaccurate.

Incorrect Answers

A:

Losing usage metrics undermines the requested allocation, which does not satisfy the requirement to measure inference cost by workload rather than averaging all project charges. A suitable test of the correct implementation is: Reconcile sample requests against measured prompt/output tokens and project cost records. This catches the particular incident where finance cannot reconcile a large token-spend increase to any application.

B:

Agents have different token and tool usage profiles. This shortcut can leave the following observable failure unresolved: Finance cannot reconcile a large token-spend increase to any application. A proper deployment instead must measure inference cost by workload rather than averaging all project charges; the missing capability would leave the deployment exposed to finance cannot reconcile a large token-spend increase to any application. The right approach is to tag deployments and correlate per-request model/token telemetry with workload identity and billing data, rather than relying on this unrelated shortcut. The distinguishing test is: Reconcile sample requests against measured prompt/output tokens and project cost records.

D:

Screenshots cannot reconstruct cost per request, so it cannot establish measure inference cost by workload rather than averaging all project charges. The working design instead needs to tag deployments and correlate per-request model/token telemetry with workload identity and billing data. A strong boundary test is: Check that tool and retry costs are counted without double charging failed calls. This matters because cross-service tool charges and cached-input discounts can make simplistic per-token estimates inaccurate.

Question 61

A Foundry application has a specific incident: Finance cannot reconcile a large token-spend increase to any application. The operator is trying to measure inference cost by workload rather than averaging all project charges. Which failure mechanism best explains the evidence?

  1. The deployed managed identity lacks a resource role or the SDK uses the wrong identity.
  2. Traces omit intermediate tool result and correlation identifiers.
  3. Indexer extraction or source synchronization failed for a subset of documents.
  4. Request traces or usage records lack reliable workload attribution.

Explanation

Correct Answer

D:

The failure that finance cannot reconcile a large token-spend increase to any application points to request traces or usage records lack reliable workload attribution. The direct functional check is: Reconcile sample requests against measured prompt/output tokens and project cost records. Then test the independent boundary—check that tool and retry costs are counted without double charging failed calls—to distinguish a genuine repair from a passing routine response.

Incorrect Answers

A:

If the deployed managed identity lacks a resource role or the SDK uses the wrong identity, then the expected symptom is inference returns an authorization error after moving from local development to the container; this incident instead reports that finance cannot reconcile a large token-spend increase to any application. The nearer diagnosis is request traces or usage records lack reliable workload attribution; to distinguish it from the alternative of the deployed managed identity lacks a resource role or the SDK uses the wrong identity, test the specific service behavior: Reconcile sample requests against measured prompt/output tokens and project cost records.

B:

If traces omit intermediate tool result and correlation identifiers, then the expected symptom is operators cannot tell whether the calculator tool or the model invented a final total; this incident instead reports that finance cannot reconcile a large token-spend increase to any application. The nearer diagnosis is request traces or usage records lack reliable workload attribution; to distinguish it from the alternative of traces omit intermediate tool result and correlation identifiers, test the specific service behavior: Reconcile sample requests against measured prompt/output tokens and project cost records.

C:

If indexer extraction or source synchronization failed for a subset of documents, then the expected symptom is some updated contracts never appear in search results even though queries succeed; this incident instead reports that finance cannot reconcile a large token-spend increase to any application. The nearer diagnosis is request traces or usage records lack reliable workload attribution; to distinguish it from the alternative of indexer extraction or source synchronization failed for a subset of documents, test the specific service behavior: Reconcile sample requests against measured prompt/output tokens and project cost records.

Question 62

A reviewer must verify that the solution will measure inference cost by workload rather than averaging all project charges. The deployment context is: A product manager needs to attribute weekly agent spending to business units using separate deployments. Which TWO tests provide the strongest evidence? Select TWO answers.

  1. Count only successful model calls in the cost report while excluding tool and retry charges.
  2. Check that tool and retry costs are counted without double charging failed calls.
  3. Estimate spend from catalog unit prices without reconciling observed input and output token usage.
  4. Charge every failed attempt in full again when the same request is retried successfully.
  5. Reconcile sample requests against measured prompt/output tokens and project cost records.

Explanation

Correct Answers

B:

Check that tool and retry costs are counted without double charging failed calls exercises a relevant negative or edge-case condition. It matters here because cross-service tool charges and cached-input discounts can make simplistic per-token estimates inaccurate. The complementary positive check—reconcile sample requests against measured prompt/output tokens and project cost records—then confirms the intended business capability still works; neither observation alone settles the whole requirement.

E:

Reconcile sample requests against measured prompt/output tokens and project cost records measures the actual outcome needed to measure inference cost by workload rather than averaging all project charges. It would expose the specific failure where finance cannot reconcile a large token-spend increase to any application. Pair it with check that tool and retry costs are counted without double charging failed calls to check both successful behavior and the remaining operational boundary.

Incorrect Answers

A:

Completed-request billing must include attributable tools and retries. For this workload, the decisive functional evidence is: Reconcile sample requests against measured prompt/output tokens and project cost records. The shortcut in this option could pass even when finance cannot reconcile a large token-spend increase to any application.

C:

Static rates alone cannot prove real usage and project attribution. The deployment needs to measure inference cost by workload rather than averaging all project charges. An independent test is: Check that tool and retry costs are counted without double charging failed calls.

D:

Double counting corrupts per-workload cost totals. The real incident to guard against is that finance cannot reconcile a large token-spend increase to any application. This does not replace the case-specific validation: Check that tool and retry costs are counted without double charging failed calls.

Question 63

An Azure AI engineer must address the following production request: An enterprise RAG pipeline imports contracts nightly into Azure AI Search. The solution must detect ingestion failures before the assistant answers from incomplete content. What should be implemented?

  1. Adjust the assistant's temperature after each failed ingestion.
  2. Monitor indexer run status, source-to-index document counts, and last-success timestamps.
  3. Monitor only HTTP 200 from the search query endpoint.
  4. Use the most recent contract filename as proof every page was indexed.

Explanation

Correct Answer

B:

Freshness requires checking ingestion progress, not just endpoint availability; for this workload the decision is to monitor indexer run status, source-to-index document counts, and last-success timestamps. The acceptance result is: Compare expected contract IDs and modified timestamps with indexed documents. An independent edge-case check is: Inject a malformed document and verify ingestion alerts identify the failed item. That distinction matters because successful indexer runs do not prove extracted tables and figures retain all their meaning.

Incorrect Answers

A:

Generation settings cannot ingest documents, which does not satisfy the requirement to detect ingestion failures before the assistant answers from incomplete content. A suitable test of the correct implementation is: Compare expected contract IDs and modified timestamps with indexed documents. This catches the particular incident where some updated contracts never appear in search results even though queries succeed.

C:

A healthy query endpoint can serve stale data. This shortcut can leave the following observable failure unresolved: Some updated contracts never appear in search results even though queries succeed. A proper deployment instead must detect ingestion failures before the assistant answers from incomplete content; the missing capability would leave the deployment exposed to some updated contracts never appear in search results even though queries succeed. The right approach is to monitor indexer run status, source-to-index document counts, and last-success timestamps, rather than relying on this unrelated shortcut. The distinguishing test is: Compare expected contract IDs and modified timestamps with indexed documents.

D:

Filenames do not prove extraction completeness, so it cannot establish detect ingestion failures before the assistant answers from incomplete content. The working design instead needs to monitor indexer run status, source-to-index document counts, and last-success timestamps. A strong boundary test is: Inject a malformed document and verify ingestion alerts identify the failed item. This matters because successful indexer runs do not prove extracted tables and figures retain all their meaning.

Question 64

During incident response, telemetry shows: Some updated contracts never appear in search results even though queries succeed. The feature exists to detect ingestion failures before the assistant answers from incomplete content. What is the most relevant diagnosis?

  1. Traces omit intermediate tool result and correlation identifiers.
  2. Indexer extraction or source synchronization failed for a subset of documents.
  3. The deployed managed identity lacks a resource role or the SDK uses the wrong identity.
  4. The search dependency was not included in network isolation configuration.

Explanation

Correct Answer

B:

The failure that some updated contracts never appear in search results even though queries succeed points to indexer extraction or source synchronization failed for a subset of documents. The direct functional check is: Compare expected contract IDs and modified timestamps with indexed documents. Then test the independent boundary—inject a malformed document and verify ingestion alerts identify the failed item—to distinguish a genuine repair from a passing routine response.

Incorrect Answers

A:

If traces omit intermediate tool result and correlation identifiers, then the expected symptom is operators cannot tell whether the calculator tool or the model invented a final total; this incident instead reports that some updated contracts never appear in search results even though queries succeed. The nearer diagnosis is indexer extraction or source synchronization failed for a subset of documents; to distinguish it from the alternative of traces omit intermediate tool result and correlation identifiers, test the specific service behavior: Compare expected contract IDs and modified timestamps with indexed documents.

C:

If the deployed managed identity lacks a resource role or the SDK uses the wrong identity, then the expected symptom is inference returns an authorization error after moving from local development to the container; this incident instead reports that some updated contracts never appear in search results even though queries succeed. The nearer diagnosis is indexer extraction or source synchronization failed for a subset of documents; to distinguish it from the alternative of the deployed managed identity lacks a resource role or the SDK uses the wrong identity, test the specific service behavior: Compare expected contract IDs and modified timestamps with indexed documents.

D:

If the search dependency was not included in network isolation configuration, then the expected symptom is the model endpoint works but traffic to the search index leaves via a public address; this incident instead reports that some updated contracts never appear in search results even though queries succeed. The nearer diagnosis is indexer extraction or source synchronization failed for a subset of documents; to distinguish it from the alternative of the search dependency was not included in network isolation configuration, test the specific service behavior: Compare expected contract IDs and modified timestamps with indexed documents.

Question 65

The project team needs production acceptance evidence. Its application has this need: An enterprise RAG pipeline imports contracts nightly into Azure AI Search. The required outcome is to detect ingestion failures before the assistant answers from incomplete content. Select TWO checks that test this outcome.

  1. Compare the number of indexed items with the number of files but ignore contract version IDs.
  2. Allow a malformed extraction to be skipped silently if the overall batch job reports success.
  3. Compare expected contract IDs and modified timestamps with indexed documents.
  4. Validate only the newest contract and assume all older indexed contracts have matching modification times.
  5. Inject a malformed document and verify ingestion alerts identify the failed item.

Explanation

Correct Answers

C:

Compare expected contract IDs and modified timestamps with indexed documents measures the actual outcome needed to detect ingestion failures before the assistant answers from incomplete content. It would expose the specific failure where some updated contracts never appear in search results even though queries succeed. Pair it with inject a malformed document and verify ingestion alerts identify the failed item to check both successful behavior and the remaining operational boundary.

E:

Inject a malformed document and verify ingestion alerts identify the failed item exercises a relevant negative or edge-case condition. It matters here because successful indexer runs do not prove extracted tables and figures retain all their meaning. The complementary positive check—compare expected contract IDs and modified timestamps with indexed documents—then confirms the intended business capability still works; neither observation alone settles the whole requirement.

Incorrect Answers

A:

Counts can match while the wrong or superseded documents are indexed. For this workload, the decisive functional evidence is: Compare expected contract IDs and modified timestamps with indexed documents. The shortcut in this option could pass even when some updated contracts never appear in search results even though queries succeed.

B:

Item-level failures must be traceable to stop a stale index going unnoticed. The deployment needs to detect ingestion failures before the assistant answers from incomplete content. An independent test is: Inject a malformed document and verify ingestion alerts identify the failed item.

D:

A single fresh document cannot establish whole-corpus synchronization. The real incident to guard against is that some updated contracts never appear in search results even though queries succeed. This does not replace the case-specific validation: Inject a malformed document and verify ingestion alerts identify the failed item.

Question 66

An AI engineer is documenting caveats after implementing a solution to detect ingestion failures before the assistant answers from incomplete content. The workload is: An enterprise RAG pipeline imports contracts nightly into Azure AI Search. Which residual risk deserves explicit validation?

  1. Successful indexer runs do not prove extracted tables and figures retain all their meaning.
  2. Groundedness scores still depend on representative and correctly labeled evaluation cases.
  3. A human approval can be invalid if it does not bind the exact amount and merchant.
  4. PII masking may miss indirect identifiers and combined quasi-identifiers.

Explanation

Correct Answer

A:

Even after the team chooses to monitor indexer run status, source-to-index document counts, and last-success timestamps, successful indexer runs do not prove extracted tables and figures retain all their meaning. This matters for an enterprise RAG pipeline imports contracts nightly into Azure AI Search. The edge-case check is: Inject a malformed document and verify ingestion alerts identify the failed item. The regular success check alone—compare expected contract IDs and modified timestamps with indexed documents—cannot establish that this risk is controlled.

Incorrect Answers

B:

Groundedness scores still depend on representative and correctly labeled evaluation cases could be important if a conversational sales agent is being tested for whether it invents warranties absent from its catalog, but the actual use case is an enterprise RAG pipeline imports contracts nightly into Azure AI Search. The closest operational risk is successful indexer runs do not prove extracted tables and figures retain all their meaning; evaluate that instead of groundedness scores still depend on representative and correctly labeled evaluation cases by testing: Inject a malformed document and verify ingestion alerts identify the failed item.

C:

A human approval can be invalid if it does not bind the exact amount and merchant could be important if a travel assistant may book refundable hotels autonomously but nonrefundable purchases require manager approval, but the actual use case is an enterprise RAG pipeline imports contracts nightly into Azure AI Search. The closest operational risk is successful indexer runs do not prove extracted tables and figures retain all their meaning; evaluate that instead of a human approval can be invalid if it does not bind the exact amount and merchant by testing: Inject a malformed document and verify ingestion alerts identify the failed item.

D:

PII masking may miss indirect identifiers and combined quasi-identifiers could be important if an internal agent summarizes employee cases that include personal identifiers and protected HR notes, but the actual use case is an enterprise RAG pipeline imports contracts nightly into Azure AI Search. The closest operational risk is successful indexer runs do not prove extracted tables and figures retain all their meaning; evaluate that instead of pII masking may miss indirect identifiers and combined quasi-identifiers by testing: Inject a malformed document and verify ingestion alerts identify the failed item.

Question 67

Consider an application deployed with Microsoft Foundry: A production support agent calls several tools and occasionally produces a wrong final answer with no apparent API error. The owner wants to add end-to-end observability across model and tool steps. Which implementation is most appropriate?

  1. Record only the final customer-visible sentence.
  2. Disable tracing whenever the model invokes an external service.
  3. Trace model calls, tool invocations, arguments, timings, and correlated request IDs with safe redaction.
  4. Store unredacted access tokens in traces for easier reproduction.

Explanation

Correct Answer

C:

Structured traces reveal which step introduced bad information; for this workload the decision is to trace model calls, tool invocations, arguments, timings, and correlated request IDs with safe redaction. The acceptance result is: Reproduce a test request and inspect the corresponding model/tool spans and error paths. An independent edge-case check is: Verify credentials and private document payloads are masked in telemetry exports. That distinction matters because tracing improves diagnosis but creates a new sensitive-data retention surface.

Incorrect Answers

A:

Final output cannot identify faulty intermediate tool data, which does not satisfy the requirement to add end-to-end observability across model and tool steps. A suitable test of the correct implementation is: Reproduce a test request and inspect the corresponding model/tool spans and error paths. This catches the particular incident where operators cannot tell whether the calculator tool or the model invented a final total.

B:

Tool spans are critical for diagnosing agent behavior. This shortcut can leave the following observable failure unresolved: Operators cannot tell whether the calculator tool or the model invented a final total. A proper deployment instead must add end-to-end observability across model and tool steps; the missing capability would leave the deployment exposed to operators cannot tell whether the calculator tool or the model invented a final total. The right approach is to trace model calls, tool invocations, arguments, timings, and correlated request IDs with safe redaction, rather than relying on this unrelated shortcut. The distinguishing test is: Reproduce a test request and inspect the corresponding model/tool spans and error paths.

D:

Sensitive credentials should never enter general diagnostics, so it cannot establish add end-to-end observability across model and tool steps. The working design instead needs to trace model calls, tool invocations, arguments, timings, and correlated request IDs with safe redaction. A strong boundary test is: Verify credentials and private document payloads are masked in telemetry exports. This matters because tracing improves diagnosis but creates a new sensitive-data retention surface.

Question 68

After rollout, the team observes: Operators cannot tell whether the calculator tool or the model invented a final total. The intended design was to add end-to-end observability across model and tool steps. Which root cause should be investigated first?

  1. Missing version correlation prevents isolating model behavior from retrieval changes.
  2. The deployed managed identity lacks a resource role or the SDK uses the wrong identity.
  3. The search dependency was not included in network isolation configuration.
  4. Traces omit intermediate tool result and correlation identifiers.

Explanation

Correct Answer

D:

The failure that operators cannot tell whether the calculator tool or the model invented a final total points to traces omit intermediate tool result and correlation identifiers. The direct functional check is: Reproduce a test request and inspect the corresponding model/tool spans and error paths. Then test the independent boundary—verify credentials and private document payloads are masked in telemetry exports—to distinguish a genuine repair from a passing routine response.

Incorrect Answers

A:

If missing version correlation prevents isolating model behavior from retrieval changes, then the expected symptom is answer accuracy declines after a model and index update shipped together; this incident instead reports that operators cannot tell whether the calculator tool or the model invented a final total. The nearer diagnosis is traces omit intermediate tool result and correlation identifiers; to distinguish it from the alternative of missing version correlation prevents isolating model behavior from retrieval changes, test the specific service behavior: Reproduce a test request and inspect the corresponding model/tool spans and error paths.

B:

If the deployed managed identity lacks a resource role or the SDK uses the wrong identity, then the expected symptom is inference returns an authorization error after moving from local development to the container; this incident instead reports that operators cannot tell whether the calculator tool or the model invented a final total. The nearer diagnosis is traces omit intermediate tool result and correlation identifiers; to distinguish it from the alternative of the deployed managed identity lacks a resource role or the SDK uses the wrong identity, test the specific service behavior: Reproduce a test request and inspect the corresponding model/tool spans and error paths.

C:

If the search dependency was not included in network isolation configuration, then the expected symptom is the model endpoint works but traffic to the search index leaves via a public address; this incident instead reports that operators cannot tell whether the calculator tool or the model invented a final total. The nearer diagnosis is traces omit intermediate tool result and correlation identifiers; to distinguish it from the alternative of the search dependency was not included in network isolation configuration, test the specific service behavior: Reproduce a test request and inspect the corresponding model/tool spans and error paths.

Question 69

A release reviewer considers this scenario: A production support agent calls several tools and occasionally produces a wrong final answer with no apparent API error. The objective is to add end-to-end observability across model and tool steps. Which TWO validation outcomes directly establish that behavior? Select TWO answers.

  1. Verify tracing is enabled but never inspect a model-to-tool span for the failing request.
  2. Verify credentials and private document payloads are masked in telemetry exports.
  3. Export a trace with an embedded example API token and assume project access restrictions make the export safe.
  4. Inspect aggregate latency charts but omit per-request trace correlation and tool error fields.
  5. Reproduce a test request and inspect the corresponding model/tool spans and error paths.

Explanation

Correct Answers

B:

Verify credentials and private document payloads are masked in telemetry exports exercises a relevant negative or edge-case condition. It matters here because tracing improves diagnosis but creates a new sensitive-data retention surface. The complementary positive check—reproduce a test request and inspect the corresponding model/tool spans and error paths—then confirms the intended business capability still works; neither observation alone settles the whole requirement.

E:

Reproduce a test request and inspect the corresponding model/tool spans and error paths measures the actual outcome needed to add end-to-end observability across model and tool steps. It would expose the specific failure where operators cannot tell whether the calculator tool or the model invented a final total. Pair it with verify credentials and private document payloads are masked in telemetry exports to check both successful behavior and the remaining operational boundary.

Incorrect Answers

A:

Enabled telemetry alone cannot demonstrate end-to-end causal observability. For this workload, the decisive functional evidence is: Reproduce a test request and inspect the corresponding model/tool spans and error paths. The shortcut in this option could pass even when operators cannot tell whether the calculator tool or the model invented a final total.

C:

Secret masking has to be tested on the exported payload itself. The deployment needs to add end-to-end observability across model and tool steps. An independent test is: Verify credentials and private document payloads are masked in telemetry exports.

D:

Aggregates can hide which specific tool span failed and why. The real incident to guard against is that operators cannot tell whether the calculator tool or the model invented a final total. This does not replace the case-specific validation: Verify credentials and private document payloads are masked in telemetry exports.

Question 70

The team has implemented the selected Azure design for this scenario: A production support agent calls several tools and occasionally produces a wrong final answer with no apparent API error. It intends to add end-to-end observability across model and tool steps. Which remaining operational limitation is specifically relevant?

  1. PII masking may miss indirect identifiers and combined quasi-identifiers.
  2. Groundedness scores still depend on representative and correctly labeled evaluation cases.
  3. Detailed traces can leak patient data if retention and access are not restricted.
  4. Tracing improves diagnosis but creates a new sensitive-data retention surface.

Explanation

Correct Answer

D:

Even after the team chooses to trace model calls, tool invocations, arguments, timings, and correlated request IDs with safe redaction, tracing improves diagnosis but creates a new sensitive-data retention surface. This matters for a production support agent calls several tools and occasionally produces a wrong final answer with no apparent API error. The edge-case check is: Verify credentials and private document payloads are masked in telemetry exports. The regular success check alone—reproduce a test request and inspect the corresponding model/tool spans and error paths—cannot establish that this risk is controlled.

Incorrect Answers

A:

PII masking may miss indirect identifiers and combined quasi-identifiers could be important if an internal agent summarizes employee cases that include personal identifiers and protected HR notes, but the actual use case is a production support agent calls several tools and occasionally produces a wrong final answer with no apparent API error. The closest operational risk is tracing improves diagnosis but creates a new sensitive-data retention surface; evaluate that instead of pII masking may miss indirect identifiers and combined quasi-identifiers by testing: Verify credentials and private document payloads are masked in telemetry exports.

B:

Groundedness scores still depend on representative and correctly labeled evaluation cases could be important if a conversational sales agent is being tested for whether it invents warranties absent from its catalog, but the actual use case is a production support agent calls several tools and occasionally produces a wrong final answer with no apparent API error. The closest operational risk is tracing improves diagnosis but creates a new sensitive-data retention surface; evaluate that instead of groundedness scores still depend on representative and correctly labeled evaluation cases by testing: Verify credentials and private document payloads are masked in telemetry exports.

C:

Detailed traces can leak patient data if retention and access are not restricted could be important if a healthcare appointment agent must explain why a scheduling tool call was executed during an audit, but the actual use case is a production support agent calls several tools and occasionally produces a wrong final answer with no apparent API error. The closest operational risk is tracing improves diagnosis but creates a new sensitive-data retention surface; evaluate that instead of detailed traces can leak patient data if retention and access are not restricted by testing: Verify credentials and private document payloads are masked in telemetry exports.

Developing AI Apps and Agents on Azure (AI-103) Practice Tests

SPECIAL OFFER: GET 10% OFF
This is ONE TIME OFFER

You save
10%

Enter Your Email Address to Receive Your 10% Off Discount Code

SPECIAL OFFER: GET 10% OFF

You save
10%

Use Discount Code:

A confirmation link was sent to your e-mail.

Please check your mailbox for a message from support@examlabs.com and follow the directions.

Download Free Demo of VCE Exam Simulator

Experience Avanset VCE Exam Simulator for yourself.

Simply submit your email address below to get started with our interactive software demo of your free trial.

  • Realistic exam simulation and exam editor with preview functions
  • Whole exam in a single file with several different question types
  • Customizable exam-taking mode & detailed score reports