McAfee Secured

Exam Code: DP-800

Exam Name: Developing AI-Enabled Database Solutions

Certification Provider: Microsoft

Corresponding Certification: Microsoft Certified: SQL AI Developer Associate

100% Microsoft DP-800 Guaranteed Success With Testking Training Materials

Get Real DP-800 Exam Questions & Expert Verified Answers as Experienced on the Real Exam!

101 Questions & Answers with Testing Engine

"Developing AI-Enabled Database Solutions Exam", also known as DP-800 exam, is a Microsoft certification exam.

Pass your tests with the always up-to-date DP-800 Exam Engine. Your DP-800 training materials keep you at the head of the pack!

Satisfaction Guaranteed

Money Back Guarantee

Testking provides hassle-free money back guarantee with our products. That is because we have 100% trust in the abilities of our professional and experience product team, and our record is a proof of that.

99.6% PASS RATE
Was: $137.49
Now: $124.99

Microsoft DP-800 Practice Test 67

Question 660

A tenant chatbot searches a shared SQL knowledge base containing documents from several customers. Security requires every response to exclude chunks that the signed-in tenant is not permitted to read, even when those chunks are highly relevant. Which control should be implemented?

  1. Retrieve only the sources the caller is authorized to access
  2. Guard retrieved documents against prompt-injection instructions
  3. Use retrieval-augmented generation for answers grounded in changing governed data
  4. Require a no-answer fallback when retrieved evidence is insufficient

Explanation

Correct Answer

A:

This approach prevents RAG from bypassing document-level permissions.

Incorrect Answers

B:

This alternative treats database content as data rather than high-priority commands. This does not meet the need for a method that prevents RAG from bypassing document-level permissions.

C:

This option injects authorized current context into prompts before generation. This does not meet the need for a method that prevents RAG from bypassing document-level permissions.

D:

This option avoids unsupported fabricated claims. This does not meet the need for a method that prevents RAG from bypassing document-level permissions.

Question 661

A RAG system sends an entire 200-page manual for a one-sentence question. What is the most appropriate next design step?

  1. Limit retrieved context to relevant chunks within token budget
  2. Guard retrieved documents against prompt-injection instructions
  3. Require a no-answer fallback when retrieved evidence is insufficient
  4. Use retrieval-augmented generation for answers grounded in changing governed data

Explanation

Correct Answer

A:

Limit retrieved context to relevant chunks within token budget is the appropriate choice: it prevents excessive noise and unnecessary model cost. The deployment should be tested with representative data.

Incorrect Answers

B:

That technique treats database content as data rather than high-priority commands. This does not meet the need for a method that prevents excessive noise and unnecessary model cost.

C:

This option avoids unsupported fabricated claims. This does not meet the need for a method that prevents excessive noise and unnecessary model cost.

D:

That technique injects authorized current context into prompts before generation. This does not meet the need for a method that prevents excessive noise and unnecessary model cost.

Question 662

A long troubleshooting question combines many retrieved SQL rows into one oversized prompt. What should the SQL developer implement?

  1. Test prompt and context size limits on worst-case queries
  2. Measure retrieval recall separately from generation quality
  3. Evaluate answer faithfulness against retrieved source passages
  4. Evaluate refusal and escalation on sensitive unsupported requests

Explanation

Correct Answer

A:

This implementation prevents truncation and silent context loss.

Incorrect Answers

B:

This choice identifies whether failures originate in search or model reasoning. This does not meet the need for a method that prevents truncation and silent context loss.

C:

This choice detects claims unsupported by context even when fluent. This does not meet the need for a method that prevents truncation and silent context loss.

D:

This choice reduces high-impact unsupported guidance. This does not meet the need for a method that prevents truncation and silent context loss.

Question 663

A legal assistant returns a summary without citation to the governing policy. Which change most directly meets the requirement?

  1. Use retrieval-augmented generation for answers grounded in changing governed data
  2. Require a no-answer fallback when retrieved evidence is insufficient
  3. Include source identifiers and relevant excerpts with generated answers
  4. Retrieve only the sources the caller is authorized to access

Explanation

Correct Answer

C:

Include source identifiers and relevant excerpts with generated answers is the appropriate choice: it makes claims traceable to retrieved passages. The deployment should be tested with representative data.

Incorrect Answers

A:

This alternative injects authorized current context into prompts before generation. This does not meet the need for a method that makes claims traceable to retrieved passages.

B:

This alternative avoids unsupported fabricated claims. This does not meet the need for a method that makes claims traceable to retrieved passages.

D:

This choice prevents RAG from bypassing document-level permissions. This does not meet the need for a method that makes claims traceable to retrieved passages.

Question 664

A REST model provider occasionally times out and open SQL sessions accumulate. The SQL deployment supports the named external model and REST integration. What is the most appropriate next design step?

  1. Convert selected SQL rows to a compact JSON payload for model context
  2. Set bounded timeout and retry behavior for SQL-external calls
  3. Move slow model calls outside long-running SQL transactions
  4. Use managed identity or approved credential isolation for external model calls

Explanation

Correct Answer

B:

Set bounded timeout and retry behavior for SQL-external calls is the appropriate choice: it protects database resources when models are slow or unavailable. The deployment should be tested with representative data.

Incorrect Answers

A:

That technique structures authorized database facts in a model-consumable form. This does not meet the need for a method that protects database resources when models are slow or unavailable.

C:

That technique avoids holding row locks while waiting for network inference. This does not meet the need for a method that protects database resources when models are slow or unavailable.

D:

That technique avoids plaintext secrets in stored procedures and deployment scripts. This does not meet the need for a method that protects database resources when models are slow or unavailable.

Question 665

An AI response must populate a fixed result schema in SQL. What should the engineer choose?

  1. Convert selected SQL rows to a compact JSON payload for model context
  2. Use managed identity or approved credential isolation for external model calls
  3. Validate and parse returned model JSON before downstream database writes
  4. Set bounded timeout and retry behavior for SQL-external calls

Explanation

Correct Answer

C:

Validate and parse returned model JSON before downstream database writes fits because it prevents malformed or invented model fields from corrupting data. That addresses the stated technical constraint.

Incorrect Answers

A:

This choice structures authorized database facts in a model-consumable form. This does not meet the need for a method that prevents malformed or invented model fields from corrupting data.

B:

This alternative avoids plaintext secrets in stored procedures and deployment scripts. This does not meet the need for a method that prevents malformed or invented model fields from corrupting data.

D:

That technique protects database resources when models are slow or unavailable. This does not meet the need for a method that prevents malformed or invented model fields from corrupting data.

Question 666

A checkout procedure starts an external model request while holding row locks on reserved inventory, so transactions stall on slow inference. Which structural change most directly reduces the lock duration?

  1. Convert selected SQL rows to a compact JSON payload for model context
  2. Invoke an approved external REST endpoint from supported SQL using sp_invoke_external_rest_endpoint
  3. Use managed identity or approved credential isolation for external model calls
  4. Move slow model calls outside long-running SQL transactions

Explanation

Correct Answer

D:

Move slow model calls outside long-running SQL transactions is the appropriate choice: it avoids holding row locks while waiting for network inference. The deployment should be tested with representative data.

Incorrect Answers

A:

This option structures authorized database facts in a model-consumable form. This does not meet the need for a method that avoids holding row locks while waiting for network inference.

B:

This choice integrates supported SQL workloads with external model APIs. This does not meet the need for a method that avoids holding row locks while waiting for network inference.

C:

This choice avoids plaintext secrets in stored procedures and deployment scripts. This does not meet the need for a method that avoids holding row locks while waiting for network inference.

Question 667

An approved policy repository has no rule addressing a novel exception request. What should the engineer choose?

  1. Use retrieval-augmented generation for answers grounded in changing governed data
  2. Require a no-answer fallback when retrieved evidence is insufficient
  3. Retrieve only the sources the caller is authorized to access
  4. Limit retrieved context to relevant chunks within token budget

Explanation

Correct Answer

B:

This approach avoids unsupported fabricated claims.

Incorrect Answers

A:

This alternative injects authorized current context into prompts before generation. This does not meet the need for a method that avoids unsupported fabricated claims.

C:

That technique prevents RAG from bypassing document-level permissions. This does not meet the need for a method that avoids unsupported fabricated claims.

D:

This alternative prevents excessive noise and unnecessary model cost. This does not meet the need for a method that avoids unsupported fabricated claims.

Question 668

An Azure SQL stored procedure needs to call an authorized AI endpoint with a bounded request. The SQL deployment supports the named external model and REST integration. Which technical choice addresses this need?

  1. Invoke an approved external REST endpoint from supported SQL using sp_invoke_external_rest_endpoint
  2. Set bounded timeout and retry behavior for SQL-external calls
  3. Use managed identity or approved credential isolation for external model calls
  4. Move slow model calls outside long-running SQL transactions

Explanation

Correct Answer

A:

This approach integrates supported SQL workloads with external model APIs.

Incorrect Answers

B:

This choice protects database resources when models are slow or unavailable. This does not meet the need for a method that integrates supported SQL workloads with external model APIs.

C:

This alternative avoids plaintext secrets in stored procedures and deployment scripts. This does not meet the need for a method that integrates supported SQL workloads with external model APIs.

D:

This choice avoids holding row locks while waiting for network inference. This does not meet the need for a method that integrates supported SQL workloads with external model APIs.

Question 669

A policy assistant must answer using rules updated weekly without retraining each time. Which approach would you recommend?

  1. Include source identifiers and relevant excerpts with generated answers
  2. Limit retrieved context to relevant chunks within token budget
  3. Use retrieval-augmented generation for answers grounded in changing governed data
  4. Retrieve only the sources the caller is authorized to access

Explanation

Correct Answer

C:

This implementation fits because it injects authorized current context into prompts before generation. That addresses the stated technical constraint.

Incorrect Answers

A:

This alternative makes claims traceable to retrieved passages. This does not meet the need for a method that injects authorized current context into prompts before generation.

B:

This alternative prevents excessive noise and unnecessary model cost. This does not meet the need for a method that injects authorized current context into prompts before generation.

D:

This alternative prevents RAG from bypassing document-level permissions. This does not meet the need for a method that injects authorized current context into prompts before generation.

Question 670

A stored help article contains text telling the assistant to ignore security rules. What is the most appropriate design decision?

  1. Limit retrieved context to relevant chunks within token budget
  2. Guard retrieved documents against prompt-injection instructions
  3. Include source identifiers and relevant excerpts with generated answers
  4. Retrieve only the sources the caller is authorized to access

Explanation

Correct Answer

B:

Guard retrieved documents against prompt-injection instructions fits because it treats database content as data rather than high-priority commands. That addresses the stated technical constraint.

Incorrect Answers

A:

This alternative prevents excessive noise and unnecessary model cost. This does not meet the need for a method that treats database content as data rather than high-priority commands.

C:

This option makes claims traceable to retrieved passages. This does not meet the need for a method that treats database content as data rather than high-priority commands.

D:

That technique prevents RAG from bypassing document-level permissions. This does not meet the need for a method that treats database content as data rather than high-priority commands.

Developing AI-Enabled Database Solutions (DP-800) Practice Tests