Microsoft Certified: SQL AI Developer Associate — DP-800
Complete Study Course
Exam: DP-800, Developing AI-Enabled Database Solutions
Audience: Database developers who design, secure, deploy and add AI capabilities to solutions on SQL Server 2025, Azure SQL and SQL database in Microsoft Fabric. You should write T-SQL confidently, know CI/CD with GitHub, and understand embeddings, vectors and models.
Domain Blueprint
| Domain | Weight | Lessons | Focus |
|---|---|---|---|
| 1. Design and develop database solutions | 35–40% | 8 | Tables and indexes; temporal, ledger, in-memory, external and graph tables; JSON; constraints, sequences and partitioning; programmability and errors; CTEs, window functions, regex and fuzzy matching; Copilot and MCP |
| 2. Secure, optimize, and deploy database solutions | 35–40% | 8 | Always Encrypted; DDM, RLS and permissions; passwordless access and auditing; isolation; Query Store and deadlocks; SQL Database Projects CI/CD; Data API builder; change handling |
| 3. Implement AI capabilities in database solutions | 25–30% | 6 | External models; embeddings and chunking; full-text, vector and hybrid search; vector indexes (ANN vs ENN); RRF; RAG with sp_invoke_external_rest_endpoint |
On Skillmarked: 22 free lessons and 22 practice questions, with a practice scenario after every lesson.
Newer features: DP-800 covers recent SQL features (vector indexes, regular expressions, fuzzy matching, change event streaming and others). Several are generally available in Azure SQL Database and SQL database in Fabric but still in preview in SQL Server 2025. The lessons note the status where it matters.
Recommended Study Plan (30–40 hours with T-SQL experience)
Week 1 – Modern T-SQL (Domain 1)
- Build tables with the json and vector types, temporal and ledger tables, and a CHECK constraint that uses REGEXP_LIKE
- Write CTEs, window functions (LAG, ROW_NUMBER), OPENJSON shredding and a graph MATCH query
- Add a
.github/copilot-instructions.mdfile and try Copilot with a read-only MCP connection
Week 2 – Production-ready (Domain 2)
- Configure Always Encrypted, RLS with SESSION_CONTEXT and a block predicate, and dynamic data masking
- Build an SDK-style SQL project, deploy it with SqlPackage from GitHub Actions using OIDC, and run DriftReport
- Expose a table and a stored procedure with Data API builder; reproduce a deadlock and read it
Week 3 – AI in the database (Domain 3)
- Create an EXTERNAL MODEL, chunk and embed documents, and compare exact and approximate vector search
- Implement hybrid search with RRF and a small RAG flow with sp_invoke_external_rest_endpoint
- Take timed mock exams and go back over your weakest domain
Critical Concepts (High-Yield)
Who must not see the data? TDE covers stolen files. Always Encrypted covers DBAs (deterministic for equality, enclaves for ranges). DDM disguises values but isn't a boundary.
Vectors: vector(n) with dimensions that match the model; VECTOR_DISTANCE is exact (ENN, about 50,000 candidates or fewer); a DiskANN index + VECTOR_SEARCH ... TOP (N) WITH APPROXIMATE is approximate (ANN). Index and query use the same metric.
Change handling: Change Tracking shows which rows, CDC shows full images, CES pushes events, the Azure Functions SQL trigger runs async reactions. Never call models from synchronous triggers.
CI/CD: Build validates and creates a .dacpac, Publish applies differences, DriftReport catches out-of-band changes, and post-deployment MERGE handles reference data.