DP-800 prerequisites: what you need first
DP-800 has no formal prerequisite: Microsoft does not require another certification before you book it. What it does assume is experience. The audience profile expects you to write T-SQL, develop databases on Microsoft SQL platforms, be familiar with CI/CD practices in GitHub and AI-assisted development tools, and understand AI concepts such as embeddings, vectors and models. If any of those are missing, fill the gap before you start the study plan.
What Microsoft expects
| Area | What you should be able to do |
|---|---|
| T-SQL | Write joins, aggregates, CTEs, window functions, procedures and error handling without looking up the syntax |
| Database design | Choose data types, keys, indexes and constraints; read an execution plan |
| Security | Explain permissions, roles, encryption and auditing at a basic level |
| CI/CD | Use Git branches, pull requests and a pipeline that builds and deploys |
| AI concepts | Explain what an embedding is, what a language model does and why prompts matter |
Skills to check yourself on
- Write a window function from memory. If
ROW_NUMBER() OVER (PARTITION BY ... ORDER BY ...)takes you minutes, practise first. - Read JSON in T-SQL.
OPENJSONandJSON_VALUEappear throughout, including in the RAG section. - Use Git. Create a branch, open a pull request and resolve a conflict. The CI/CD objectives assume this.
- Know what a transaction does. Isolation levels, blocking and deadlocks only make sense if you understand locks and rollbacks.
- Call a REST API. RAG in T-SQL means sending an HTTPS request with a JSON body and reading the JSON that comes back.
Platforms you should have touched
The exam covers SQL Server, Azure SQL and SQL databases in Microsoft Fabric. You do not need production experience on all three, but run labs on at least one. A free Fabric trial or a small Azure SQL database is enough. Note that some newer features, such as the vector type and the regular expression functions, depend on platform and version.
What you do not need
- You do not need to be a machine learning engineer. You do not train models; you call them and store what they return.
- You do not need Python. The AI work happens in T-SQL, with Azure Functions or Logic Apps as optional glue.
- You do not need an architect-level knowledge of Azure; a working knowledge of managed identities and Azure Monitor is enough.
If you are coming from another exam
| Background | What carries over | What to add |
|---|---|---|
| DP-700 Fabric data engineer | SQL, Fabric, OneLake | Row-level security in SQL, database projects, vector search |
| AI-103 AI apps and agents | Embeddings, RAG ideas | T-SQL depth, database security, query tuning |
| DBA background | Performance, security, backups | CI/CD, JSON, Data API builder, AI search |
Next, read the exam format and then pick your study route. For a view of the whole skill list in one place, see the skills outline guide.