Generative AI Engineer Associate exam guide
Who the Generative AI Engineer Associate exam is for
According to its exam guide, the Databricks Certified Generative AI Engineer Associate exam assesses the ability to design and implement LLM-enabled solutions using Databricks. That includes breaking complex requirements into manageable tasks and choosing models, tools and approaches for a complete solution. It also covers Databricks-specific tools: Vector Search for semantic similarity search, Model Serving for deployment, MLflow for the solution lifecycle and Unity Catalog for data governance.
The guide expects successful candidates to build and deploy RAG applications and LLM chains on the platform. No prerequisite is required; related training and six months of hands-on experience are recommended. Expect questions that start from a requirement and ask for the right chunking strategy, model, chain component, serving option or monitoring feature.
Exam format
| Parameter | Value |
|---|---|
| Scored questions | 45 |
| Question types | Multiple choice and multiple selection (one or several answers) |
| Time limit | 90 minutes |
| Unscored questions | Up to 10 may be included; not identified, no effect on the score, extra time already counted |
| Delivery | Online or test center |
| Test aids | None allowed |
| Languages | English, Japanese, Brazilian Portuguese, Korean |
| Passing score | Not published on the exam page or in the exam guide |
| Validity | 2 years; to recertify, you take the full exam that is live at that time |
| Current exam version | Published March 18, 2026 |
Because unscored questions are not identified, treat every question as if it counts and keep an even pace.
Sections and weights
The exam guide splits the content into six sections, each with a list of objectives.
| Section | Weight | Main topics |
|---|---|---|
| Design Applications | 14 % | Prompts that return a set format, model tasks for a business need, chain components, tools for multi-stage reasoning, Agent Bricks |
| Data Preparation | 14 % | Chunking strategies, filtering noisy content, Python packages for document extraction, writing chunks to Delta Lake tables in Unity Catalog, retrieval metrics, re-ranking |
| Application Development | 30 % | LangChain and similar tools, prompt augmentation, guardrails, choosing an LLM and an embedding model, model cards, MLflow and Agent Framework, Genie for data retrieval |
| Assembling and Deploying Applications | 22 % | pyfunc chains, access control from serving endpoints, registering models in Unity Catalog, Vector Search indexes, Foundation Model APIs, ai_query() for batch inference, CI/CD, MCP servers, prompt versioning |
| Governance | 8 % | Masking, guardrails against malicious input, licensing of data sources, handling problematic text |
| Evaluation and Monitoring | 12 % | Metrics for choosing and monitoring an LLM, MLflow scoring and tracing, inference tables, AI Gateway, cost control, custom scorers, expert feedback |
Application Development and Assembling and Deploying Applications carry 52 % of the weight between them, and they hold the longest lists of objectives. Governance is the smallest section; guardrails also appear among the Application Development objectives.
Product names that changed since the guide
The exam guide still uses product names from the March 2026 exam version, while the Databricks documentation has renamed or repositioned several of them. The guide itself tells candidates to study the documentation to learn current, non-deprecated features. Know both names, so that a question written with the guide's terms does not surprise you:
- Vector Search: the documentation states that Databricks AI Search was formerly known as Databricks Vector Search.
- Genie Spaces: the documentation states that Genie Agents were formerly known as Genie Spaces.
- Agents on Model Serving: Databricks now recommends authoring agents on Databricks Apps rather than on Model Serving.
How to prepare
- Read the exam page for registration and current exam conditions.
- Read the exam guide and copy every objective into a checklist. The guide asks you to check back two weeks before your exam for a newer version.
- Take the instructor-led Generative AI Engineering with Databricks course, or the self-paced Databricks Academy courses the guide lists: Building RAG Agents with Agent Bricks, Building Agentic Applications on Databricks, Agent Evaluation on Databricks, and Deploying and Monitoring Agent Applications on Databricks.
- Build one small RAG application end to end in Python: extract and chunk documents, write them to a Delta table, create a search index, assemble a chain, register it with MLflow and serve it. This covers a good part of the Data Preparation and Assembling and Deploying objectives.
- Spend extra time on the two heaviest sections, and for each choice the guide mentions (chunking, embedding context length, model size) note the criteria that decide it.
- Learn the evaluation and monitoring vocabulary: judges that need ground truth, tracing, inference tables, rate limiting and cost controls.
- Practise with our Generative AI Engineer Associate practice exam under timed conditions, then review each missed question in the free study mode.
Practice Exam: Certified Generative AI Engineer Associate
25 free questions with explanations, no account needed.