Databricks Certified Generative AI Engineer Associate

Covers designing and deploying LLM applications on Databricks: RAG, chains, agents, serving and evaluation.

What it signals

Practical assessment: Scenario-based multiple choice. Proctored multiple-choice with code-reading items, no labs.
Identity assurance: Proctored with identity verification. Proctored; online proctored or test centre (Kryterion/Webassessor). No test aides.
Depth: Associate. Associate level.
Experience expected: Under 6 months. Provider recommends 6+ months hands-on experience.
Portability: Mixed. Concepts (RAG, chunking, guardrails, MCP) transfer, but many items are Databricks-tool specific.
Verifiability: Public badge or registry. Public Accredible badge URL and optional certified directory.

For hiring managers

The March 2026 blueprint added substantial agent content (Agent Bricks, MCP servers, multi-agent with Genie, agent evaluation with MLflow tracing, prompt versioning), so a pass before 2026 reflects a more RAG-centric exam. Passing score is undisclosed. Retakes: 14-day wait, full price each time.

Verifying it

Digital badge issued via Accredible at credentials.databricks.com (shareable URL an employer can open); holders may also opt in to the Databricks Certified Directory.

What it covers

Published exam outline

Design Applications 14%
Data Preparation 14%
Application Development 30%
Assembling and Deploying Apps 22%
Governance 8%
Evaluation and Monitoring 12%

By topic

GenAI apps: Core focus (over 35%). Prompting, RAG, model selection and evaluation dominate the blueprint (Design 14% + App Development 30% + much of Deploy/Eval).
Agents: Some (10 to 20%). Agent objectives (Agent Bricks, MCP, multi-agent, agent evaluation) appear across several sections but have no dedicated weight; estimated 10-20%.
Classical ML: Not covered. No model training or tuning objectives.
Operations: Substantial (20 to 35%). Assembling/Deploying 22% plus Evaluation and Monitoring 12% cover serving, CI/CD and monitoring.
Data: Some (10 to 20%). Data Preparation is 14% (chunking, extraction, Delta tables).
Responsible AI: Touched on (under 10%). Governance section is 8%.

Exam details

Who the provider says it is for. Individuals who design and implement LLM-enabled solutions on Databricks (RAG applications, LLM chains, agents) using Vector Search, Model Serving, MLflow and Unity Catalog.

Format: Multiple choice / multiple selection (45 scored; unscored items may be added)
Delivery: Proctored; online proctored or test centre (Kryterion/Webassessor). No test aides.
Code or hands-on work: Candidates read code in multiple-choice items (Python; SQL for data manipulation). No code writing or labs.
Prerequisites: None
Recommended experience: 6+ months hands-on experience with generative AI solution tasks; related training recommended
Renewal: Retake the full current exam every 2 years

Preparing

Not confirmed

These details were not available from an official source at the time of research.

passingScore (not published by Databricks)
examCode (Databricks does not use exam codes)
Exact share of agent-related questions (agent objectives are spread across sections; no separate weight)

Sources (3)