Azure Databases
AI-200 data domain · optional AZ-305 architect track

Databases forAI appsand architects

From your first query to vector search and tuning with Azure Cosmos DB for NoSQL and Azure Database for PostgreSQL. An optional architect track adds SQL and data lake design.

Begin the journey Jump to the engines
11visual
explainers
2engines with
vector search
MS Learnevery fact
grounded
Start here

Meet the team, then pick the right database

A short intro to the journey, then one orientation that matches the workload to the engine before you write a query.

00
Journey intro

One database cannot do it all

Meet the specialist team behind Contoso’s AI support assistant, see the route through the journey, and learn how the developer and architect tracks fit together.

Start the intro →
01
Cosmos DB
vs
PostgreSQL
Orientation

Pick your database for AI apps

Compare Azure Cosmos DB for NoSQL and Azure Database for PostgreSQL for AI workloads, so you know which track to follow and why.

Start step 01 →
Main path

AI developer track · Cosmos DB and PostgreSQL (AI-200)

Each engine follows the same arc. Dive down the cylinder: build and query, add vector search, then optimize. Take one engine, or both side by side.

Build & query→Vector search→Optimize

Azure Cosmos DB for NoSQL

documents · JSON · request units

  1. 02
    Foundation

    Cosmos DB: build and query

    Connect with the SDK, work with items, and write SQL queries that read documents for AI apps.

    Start step 02 →
  2. 03
    Vectors

    Cosmos DB: vector search

    Store embeddings, rank results with VectorDistance, add metadata filters and hybrid search, and keep embeddings in sync with the change feed.

    Start step 03 →
  3. 04
    Optimize

    Cosmos DB: optimize performance

    Analyze query patterns, configure range and composite indexes, pick a vector index type, and choose a consistency level that balances freshness and cost.

    Start step 04 →
Cosmos DB learning path ↗

Azure Database for PostgreSQL

relational · SQL · pgvector

  1. 05
    Foundation

    PostgreSQL: build and query

    Design schemas, write efficient SQL, and connect from Python with secure Microsoft Entra authentication.

    Start step 05 →
  2. 06
    Vectors

    PostgreSQL: vector search with pgvector

    Store embeddings with pgvector, compare distance metrics, and build retrieval patterns for RAG, semantic search and recommendations.

    Start step 06 →
  3. 07
    Optimize

    PostgreSQL: optimize vector search

    Tune PostgreSQL and pgvector, choose vector indexes, shape the data layout, scale for volume, and pool connections.

    Start step 07 →
PostgreSQL learning path ↗
Put it all together · Capstone

Contoso ships the assistant

Follow one customer question through chat history, vector search, an orders lookup and tuning, and see how every engine you learned plays its part.

Start step 08
AI-200 alignment

What AI-200 expects from these databases

The AI-200 study guide asks you to be comfortable with the skills below. The engine stops above walk through them in a visual, hands-off way.

Azure Cosmos DB for NoSQL

  • Use the SDK and write queries
  • Set the indexing policy and consistency level for efficient RU use
  • Store embeddings and run vector search
  • React to changes with the change feed processor

Azure Database for PostgreSQL

  • Use the SDK and write queries
  • Design schemas and indexes
  • Measure and reduce pgvector latency
  • Size compute, memory and storage for vector workloads
  • Build RAG with metadata filters
  • Optimize connections

Not covered yet: Azure Managed Redis is also on the AI-200 study guide, but it is not part of this journey yet.

Read the AI-200 study guide ↗
Optional · outside the AI-200 path

Architect track (AZ-305)

Not needed for the AI developer path. Take these if you design data platforms and want the architect view of relational data and data integration.

09
Architect track Optional

Design relational data storage

Choose between SQL Database, SQL Managed Instance and SQL Server on Azure VMs, then plan for scalability, availability and security.

Start step 09 →
10
Architect track Optional

Design data integration and data lake

Compare Data Factory, Data Lake, Databricks and Synapse, plan hot, warm and cold paths, and add Stream Analytics for streaming data.

Start step 10 →
Recap

Four stops that matter

Whichever engine you pick, the path is the same: choose it on purpose, build a solid foundation, add vectors for AI, then tune for cost and speed. The architect track is there when you need the wider data platform.

  1. PickMatch the workload to Cosmos DB or PostgreSQL before you build.
  2. Build and queryConnect with the SDK, model your data, and write efficient queries.
  3. Add vectorsStore embeddings and run similarity search with metadata filters.
  4. OptimizeTune indexes, consistency and connections for latency and cost.
Go deeper on Microsoft Learn

Official hands-on training

Two guided learning paths for the AI developer track, and two modules for the optional architect track.

Cosmos DB learning path ↗ PostgreSQL learning path ↗ AZ-305: relational data ↗ AZ-305: data integration ↗