Study Notes

Appearance

DP-800: Exam Guide ​

← Overview · Cheatsheet →

How the Exam Wants You to Think ​

DP-800 is not a pure AI exam and not a pure DBA exam. It expects a developer mindset that balances:

  1. Correct schema and query design
  2. Production security and deployment discipline
  3. Practical AI integration using vectors, search, and RAG

If an answer is clever but unsafe, or modern but operationally weak, it is probably wrong.

Beta-Specific Facts ​

  • As of March 28, 2026, Microsoft lists DP-800 as a beta certification exam.
  • Beta exams are not scored immediately.
  • Microsoft says the Practice Assessment is not currently available while the exam remains in beta.
  • The certification page states 100 minutes and a passing score of 700.

Keyword Detection Table ​

If you see...Lean toward...
"history of changes"Temporal tables
"tamper-evident" or "prove integrity"Ledger tables
"some users should see masked values"Dynamic Data Masking
"DBA must not see plaintext"Always Encrypted
"each user sees only their own rows"Row-Level Security
"historical regressions or plan changes"Query Store
"reviewable schema deployments"SQL Database Projects
"REST and GraphQL over SQL"Data API builder
"meaning-based retrieval"Vector search
"combine keyword and semantic relevance"Hybrid search + RRF
"current enterprise data in LLM answer"RAG
"AI tool with live schema/tool context"MCP-enabled workflow

Common Traps ​

Watch these closely

  • Masking is not encryption. If the scenario is about true protection, DDM is not enough.
  • Fine-tuning is not the default. When data changes often, the stronger answer is usually retrieval plus grounding.
  • Copilot is not a control plane. AI-assisted tools still require instruction files, permissions, review, and safe pipelines.
  • Vector search is not full-text search. If the scenario is about semantic similarity, keyword indexing alone is incomplete.
  • Manual deployment scripts are not the best DevOps answer when SQL Database Projects and CI/CD are available.

Quick Decision Rules ​

"Which protection should I use?" ​

text
Need to hide values from some users -> Dynamic Data Masking
Need row-by-row visibility control -> Row-Level Security
Need strong encryption with client-side protection -> Always Encrypted
Need audit evidence -> Auditing

"Which search pattern fits?" ​

text
Exact words and operators matter most -> Full-text
Meaning and similarity matter most -> Vector
Need both business terms and semantic relevance -> Hybrid
Need generated answer grounded in retrieved data -> RAG

"Which delivery pattern fits?" ​

text
Need governed schema delivery -> SQL Database Projects + CI/CD
Need quick REST/GraphQL over SQL -> Data API builder
Need AI help writing SQL safely -> Copilot + instruction files + review

Final Strategy ​

  • Treat Domains 1 and 2 as the scoring backbone. Domain 3 is differentiating, but only after the SQL fundamentals are solid.
  • Memorize the security distinctions: Always Encrypted vs DDM vs RLS.
  • Be fluent in the search ladder: full-text -> vector -> hybrid -> RAG.
  • Expect Microsoft-preferred implementation paths: SQL Database Projects, DAB, Managed Identity, GitHub workflows, and MCP-aware tooling.

← Overview · Cheatsheet →