DP-800: SQL AI Developer Associate with vectors, RAG and security
My most recent certification combines two things I use every day, SQL and governance, with the new side of relational databases: embeddings, vector search and RAG.
by Gabriel Henrique · · 2 min read · Ler em português
I'm Gabriel Henrique, a data engineer, and in October 2026 I earned the Microsoft Certified: SQL AI Developer Associate (DP-800). It validates the ability to design, build and secure AI-enabled database solutions across SQL Server, Azure SQL and Microsoft Fabric.
SQL is the language I have used the most for years. I chose this exam because it covers exactly where SQL is heading: the relational database storing and querying vectors right next to business data.
What the certification covers
- Advanced T-SQL and JSON processing.
- Query performance tuning.
- Security: Row-Level Security, Dynamic Data Masking, Always Encrypted and managed identities.
- CI/CD with SQL database projects.
- Data API builder to expose data as an API.
- AI: embeddings, vector search and Retrieval-Augmented Generation (RAG).
Security: the part that speaks banking
I work with financial data, so the security topics were the ones I connected most to my daily work.
Row-Level Security
Filters the rows each user can see directly in the database, through a predicate function. The rule lives in one place instead of being scattered across every report.
Dynamic Data Masking
Masks sensitive columns at query time without changing the stored data. Useful in analytics environments, where the team needs the structure but not the real value.
Always Encrypted
Encrypts the column end to end. Not even the database administrator sees the plain text, because the keys stay on the client side.
Managed identities
The application authenticates to the database with its Azure identity, with no password stored in a config file.
Vector search and RAG inside the database
The core idea is to store the embedding (the vector that represents the meaning of a text) in the same table as the business record. A query can then find the items most similar to a question and still apply regular SQL filters such as date, segment and permission.
In RAG, those retrieved passages become context for a language model to answer. Doing it in the database means reusing what already exists: row-level security, backups and governance apply to the vectors as well.
CI/CD for databases
With SQL database projects, the database schema becomes code: tables, views and procedures are versioned and published through a pipeline. It's the same principle I apply on Databricks, described in the DP-750 article.
How it fits my stack
This certification closes a loop with AI Orchestration Fundamentals: one teaches how to orchestrate AI workflows, the other how to store and query the data those workflows use, securely. Both topics show up more and more in data projects.
To see my other certifications and projects, visit the portfolio of Gabriel Henrique, Data Engineer. I'm Gabriel Henrique, and I keep writing about SQL, governance and data engineering.