The short version
The Databricks Certified Generative AI Engineer Associate exam tests whether you can design, build, deploy, and evaluate LLM applications on Databricks, with retrieval-augmented generation (RAG) at the center. It has 45 scored questions and a 90-minute time limit. This guide follows the exam guide that Databricks links from its certification page, which is dated March 18, 2026 inside the document.
What the exam is
The full name is Databricks Certified Generative AI Engineer Associate. Databricks does not print an exam code for it, so the product name is the identifier. The vendor describes the exam as a check of your ability to design and implement LLM-enabled solutions on Databricks. That includes breaking a business problem into smaller tasks, and then choosing models, tools, and approaches that fit.
Four platform areas come up again and again in the outline:
- Vector Search, for semantic similarity search over your documents.
- Model Serving, for deploying models and complete applications.
- MLflow, for managing the lifecycle of a solution, including evaluation and tracing.
- Unity Catalog, for governance, access, and storing prepared data.
Databricks says people who pass can be expected to build and deploy RAG applications and LLM chains that use the platform well. In practice, most questions are short scenarios. A team has a constraint, such as cost, latency, document format, or a privacy rule, and you pick the design that fits. Some questions ask for two answers, so read the instruction at the end of each stem.
Who it is for
The audience is engineers who build generative AI applications on Databricks. Databricks requires nothing before you book, but it recommends about six months of hands-on work on the Databricks platform with the topics in the outline, and it strongly recommends related training. The machine learning code in the exam is Python. Data manipulation and workflow code that is not machine learning may appear in SQL. The prerequisites page goes through what helps.
Format at a glance
| Item | What Databricks says |
|---|---|
| Name | Databricks Certified Generative AI Engineer Associate |
| Level | Associate |
| Scored questions | 45, multiple choice and multiple selection |
| Unscored items | Up to 10 may be added; they are not marked as such |
| Time limit | 90 minutes, with time for unscored items already included |
| Passing score | Not published |
| Delivery | Online proctored per the exam guide; the certification page says online or test center |
| Languages | English, Japanese, Brazilian Portuguese, Korean |
| Test aids | None allowed |
| Validity | 2 years, then you take the current exam again |
Details on scoring, rescheduling, and retakes are on the exam format page. Fees are listed on the vendor page and change over time, so check it when you book.
The six domains and their weightings
| Domain | Weighting |
|---|---|
| Design Applications | 14% |
| Data Preparation | 14% |
| Application Development | 30% |
| Assembling and Deploying Applications | 22% |
| Governance | 8% |
| Evaluation and Monitoring | 12% |
Design Applications
Turning a business goal into inputs, outputs, and a pipeline. You write prompts that produce a specific format, pick model tasks, choose chain components, order tools for multi-step reasoning, and decide when Agent Bricks (Knowledge Assistant, Multiagent Supervisor, Information Extraction) is the right shortcut.
Data Preparation
Getting documents ready for retrieval. Expect chunking strategies, filtering out content that hurts quality, choosing a Python package to extract text from a given file type, writing chunks to Delta tables in Unity Catalog, retrieval metrics, and the role of re-ranking.
Application Development
The biggest domain. It covers prompt design, guardrails, picking an LLM or embedding model from model cards and experiment metrics, LangChain-style tools, MLflow and the Agent Framework for agentic systems, and multi-agent setups that use Genie Spaces.
Assembling and Deploying Applications
Building a chain as a pyfunc model, registering it to Unity Catalog, serving it, and controlling access from the endpoint. It also covers creating and querying Vector Search indexes, batch inference with `ai_query()`, MCP servers, prompt version control, CI/CD, and user interfaces such as Databricks Apps.
Governance
Masking, guardrails against malicious inputs, licensing and legal risk for data sources, and what to do about problematic text in a source.
Evaluation and Monitoring
Choosing metrics, MLflow scoring and tracing, judges that need ground truth, inference logging, AI Gateway, custom scorers, cost control, and using expert feedback.
Official source
For the vendor’s own page, see the Databricks certification page. It is the final authority if anything here differs, and Databricks asks you to recheck the exam guide about two weeks before your exam date.
More guides for DATABRICKS-GENAI-ENGINEER-ASSOCIATE
- Databricks GenAI Associate Study Plan
- Databricks GenAI Associate Sample Questions
- Is the Databricks GenAI Associate Hard?
- Databricks GenAI Associate Renewal and Retakes
- Is Databricks GenAI Associate Worth It?
- Databricks GenAI Associate Cheat Sheet
- GenAI vs ML Associate: Which Databricks Exam?
Next step
Read the domains explained page to see where your time should go. When you want to check yourself, our Databricks GenAI Engineer Associate practice tests are original practice sets built around these six domains.
Frequently asked questions
Does the certification expire?
Yes. It is valid for two years. To recertify, Databricks says you take the current live version of the exam.
Is there a required course?
No. Databricks requires no prerequisite, but it recommends training and lists several courses in the exam guide.
Is the exam all multiple choice?
Yes. The exam guide describes multiple-choice and multiple-selection questions, so you choose answers rather than type them. Expect to read short snippets and parameter names.