ExamSprintLab

DP-800 study plan: pass in 4 weeks

Updated October 9, 2026

Four weeks is realistic for DP-800 if you already write T-SQL at work and study 8–10 hours a week. Cover the material in this order: database objects and advanced T-SQL first, then security and performance, then CI/CD and APIs, then the AI topics last, because embeddings and RAG make sense only once you are comfortable with JSON, triggers and stored procedures. If you have never written T-SQL, add several weeks of SQL fundamentals first.

Before you start

  • Read the official skills outline once, end to end. It lists the functions, features and tools by name.
  • Set up a lab database. A SQL database in Microsoft Fabric, an Azure SQL database on a free or low tier, or a local SQL Server that supports the newest features all work. Some functions in the outline are recent, so check feature availability and compatibility level on your platform.
  • Install Visual Studio Code with the SQL extensions and GitHub Copilot, and create a GitHub repository for the CI/CD week.
  • Open the study guide and note the date of the outline you are studying against. See the exam format.

Week 1: design and develop (35–40%)

Goal: design tables and write the T-SQL the exam shows you.

  • Objects: data types and sizes, indexes and columnstore, temporal, ledger, graph, in-memory and external tables, JSON columns and indexes, constraints, sequences, partitioning.
  • Programmability: views, scalar and table-valued functions, stored procedures, triggers.
  • Advanced T-SQL: CTEs, window functions, JSON functions, regular expression functions, fuzzy matching (edit distance, Jaro-Winkler), graph MATCH queries, correlated queries, TRY...CATCH.
  • AI-assisted development: enabling Copilot, instruction files, MCP server connections, and the security risks of sending code and data to a model.

Lab: an orders schema with a temporal table, a JSON column, a sequence, a partitioned history table and a graph of customer relationships. Write one query per function family.

Week 2: secure and optimize (first half of the 35–40%)

Goal: protect the data and find out why a query is slow.

  • Security: Always Encrypted versus column-level encryption, Dynamic Data Masking, row-level security, object permissions, passwordless access, auditing, managed identity for model endpoints.
  • Performance: execution plans, DMVs, Query Store, Query Performance Insight, isolation levels, blocking and deadlocks.

Lab: add a row-level security policy driven by SESSION_CONTEXT, mask an email column, then reproduce a deadlock with two sessions and fix it. Force a plan in Query Store after a regression.

Week 3: deploy and integrate (second half of the 35–40%)

Goal: put the database under source control and expose it safely.

  • SQL Database Projects: SDK-style projects, build to a data-tier package, reference data, unit and integration tests, schema drift, publishing.
  • Pipelines: branching, pull requests, secrets management, approvals and code owners.
  • Data API builder: configuration file, entities, REST and GraphQL, caching, pagination, deployment.
  • Change handling and monitoring: change event streaming, change data capture, Change Tracking, Azure Functions SQL trigger binding, Logic Apps, Application Insights and Log Analytics.

Lab: a repository with a database project, a GitHub Actions or Azure Pipelines workflow that builds it and deploys to a test database, and a Data API builder container exposing two tables and one stored procedure.

Week 4: AI capabilities (25–30%) and review

Goal: store embeddings, search them and answer questions from the results.

  • Models and embeddings: external models, choosing and maintaining embeddings (triggers, Change Tracking, Functions, Logic Apps, CDC), which columns to embed, chunking, generating embeddings.
  • Search: full-text versus vector versus hybrid, the vector type and indexes, VECTOR_DISTANCE and VECTOR_SEARCH, ANN versus KNN, reciprocal rank fusion.
  • RAG: retrieve rows, convert them to JSON, call a language model with sp_invoke_external_rest_endpoint, extract the answer.

Lab: embed 200 product descriptions, run exact and approximate search, add full-text search, fuse the two rankings, and write a stored procedure that answers a question from the top five rows.

Review: take the free 20-question practice test, then go back to the lowest-scoring domain. Use the three domain guides in this section.

If you already develop on SQL Server or Azure SQL

Compress week 1 and half of week 2. Spend the time on the topics that appear in few day jobs: SDK-style database projects, Data API builder, MCP, external models, vector indexes, hybrid search and RAG.

How to know you are ready

  • You can read a query with window functions, JSON and a CTE and say what it returns.
  • You can say which security feature solves a given requirement, and what it does not protect against.
  • You can read a deadlock or an execution plan and name the fix.
  • You can explain why you would pick ANN over KNN, and when hybrid search beats pure vector search.
  • You score 80% or more twice in a row on realistic practice questions from different sets.

Then book it. See the exam day guide for the day itself.