The Databricks Certified Generative AI Engineer Associate is a 90-minute exam with 45 scored questions. It checks whether you can build and ship RAG apps and agents on Databricks: chunk documents, index them, wire up a chain or agent, deploy it, govern it and measure it. The fastest way to pass is to study the six sections in proportion to their weight and to build every piece yourself in a free workspace, instead of memorizing flashcards.
Below you’ll find the exam facts, the section weights, the product names that changed after the current exam guide was written, a 6-week study plan with a hands-on checklist, and a study prompt you can paste into any AI chatbot.
The exam at a glance
These details come from Databricks’ official exam page and exam guide, as of October 2026.
| Detail | What Databricks says |
|---|---|
| Scored questions | 45, multiple-choice or multiple-selection. Unscored items may be added, with extra time to cover them. |
| Time limit | 90 minutes |
| Registration fee | $200 per attempt |
| Delivery | Online proctored or at a test center |
| Languages | English, Japanese, Brazilian Portuguese, Korean |
| Prerequisites | None. Related training and 6+ months of hands-on experience are recommended. |
| Test aids | None allowed |
| Code on the exam | Python for machine learning code; SQL may appear for data manipulation |
| Passing score | Not published. Databricks says passing scores are set statistically and can change. |
| Validity | 2 years. To recertify, you take the current version of the full exam. |
What the six sections test
Databricks publishes the weights, so let them set your schedule. With 45 scored questions, 30% works out to roughly 13 or 14 questions, while 8% is about 4.
| Section | Weight | What you should be able to do |
|---|---|---|
| Design Applications | 14% | Write prompts that return a specific format, match business goals to model tasks and chain components, order tools for multi-step reasoning, and pick the right Agent Bricks option (Knowledge Assistant, the multi-agent supervisor, Information Extraction) |
| Data Preparation | 14% | Choose a chunking strategy, filter junk content, pick a Python package to extract text, write chunks to Delta tables in Unity Catalog, measure retrieval quality, explain re-ranking |
| Application Development | 30% | Choose LangChain-style tools, add guardrails, select LLMs and embedding models, build agents with MLflow and Agent Framework, compare evaluation with monitoring, let multi-agent systems pull data from Genie |
| Assembling and Deploying Apps | 22% | Code a pyfunc chain, register it to Unity Catalog with MLflow, create and query a vector index, serve with Foundation Model APIs, use ai_query() for batch jobs, connect MCP servers, version prompts, build a user-facing app |
| Governance | 8% | Mask sensitive data, defend against malicious input, respect data licensing, deal with problematic text in source data |
| Evaluation and Monitoring | 12% | Pick metrics and judges, trace and score agents in MLflow, use inference tables, usage tables and rate limits, control LLM costs, fold in feedback from subject matter experts |
Application Development plus Assembling and Deploying make up more than half the exam. If your time is short, that’s where the hours go.
Heads-up: product names changed after the guide was written
The live exam guide covers the version of the exam that was current on March 18, 2026, and Databricks renames products often. Databricks calls the exam guide the source of truth for what’s on the exam, but the docs you’ll study from use the newer names. Keep this translation table handy:
| Name in the exam guide (or older courses) | Name in current Databricks docs |
|---|---|
| Vector Search / Mosaic AI Vector Search | AI Search (the docs say “formerly Databricks Vector Search”) |
| Genie Spaces | Genie Agents (the docs say “formerly known as Genie Spaces”) |
| Multiagent Supervisor (Agent Bricks) | Supervisor Agent |
| AI Gateway for serving endpoints | Still documented but labeled legacy; the newer control plane is called Unity Gateway |
| Community Edition | Free Edition (Community Edition was retired in 2025) |
Databricks’ own AI Prep Guide warns about exactly this: chatbots trained before recent platform changes tend to use old names and describe deprecated workflows as current. Databricks also recommends checking the exam guide again about two weeks before your exam in case it has been updated.
A 6-week study plan
This plan assumes 5 to 7 hours a week and working Python. If you already build RAG apps at work, squeeze it into three or four weeks. If Databricks is new to you, stretch it to eight.
Week 1: Set up and learn the design basics
- Download the exam guide PDF and try the ten sample questions at the end before you study anything. Note the ones you miss.
- Sign up for Databricks Free Edition and Databricks Academy. Start the self-paced courses the guide names: Building Retrieval Agents on Databricks, Building Single-Agent Applications on Databricks, Generative AI Application Evaluation and Governance, and Generative AI Application Deployment and Monitoring.
- Study Design Applications: prompts that force an output format, matching a business need to a model task, and when each Agent Bricks option fits.
Week 2: Data preparation
- Parse a handful of PDFs, then try fixed-size, paragraph and header-based chunking and compare what retrieval returns for the same question.
- Write the chunks to a Delta table in Unity Catalog. Learn how chunk size, overlap and the embedding model’s token limit affect each other.
- Read the AI Search docs on hybrid search and the reranker.
Weeks 3 and 4: Application development
- Build a small RAG agent with MLflow and Agent Framework. Turn on MLflow Tracing and read a trace step by step.
- Add a guardrail, then compare two LLMs on the same set of questions using evaluation metrics.
- Learn how a supervisor system sends table questions to a Genie Agent and document questions to a retrieval agent.
Week 5: Assembling and deploying
- Register your chain to Unity Catalog under a three-level name (catalog.schema.model) and deploy it to a serving endpoint.
- Create and query an AI Search index. Know Delta Sync vs Direct Vector Access indexes, and standard vs storage-optimized endpoints.
- Run
ai_query()over a table. Read how managed, external and custom MCP servers plug into an agent, and how MLflow tracks prompt versions and promotes them with aliases.
Week 6: Governance, evaluation and review
- Mask personal data with
ai_mask(), review guardrails against malicious prompts, and read the licensing objective closely. - Run
mlflow.genai.evaluate()and learn which built-in judges need ground truth (Correctness does; Safety doesn’t). Set up inference tables and a rate limit. - Work through our 15 free Databricks practice questions, re-read the exam guide line by line, and book the exam once every objective looks familiar.
The hands-on checklist
Databricks’ AI Prep Guide is blunt: reading alone won’t pass a Databricks certification. Free Edition covers most of this list, with limits worth knowing: one AI Search endpoint, no Direct Vector Access indexes, a cap on active serving endpoints, no GPU serving, and up to three Databricks Apps.
- Load 5 to 10 PDFs into a Unity Catalog volume, extract the text, and chunk it into a Delta table.
- Create a Delta Sync index on that table. On a standard endpoint, the source table must use change data feed.
- Query the index with a metadata filter, then try hybrid search and the reranker where your workspace supports them.
- Wrap retrieval and an LLM call into a chain, log it with MLflow, and register it to Unity Catalog.
- Serve it on an endpoint and call it from a notebook.
- Turn on tracing and run an evaluation with one judge that needs ground truth and one that doesn’t.
- Summarize a text column with
ai_query()and mask names and emails withai_mask(). - Build a small Databricks App or chat front end that calls your endpoint.
Any step you can’t finish is your next study target.
Use AI as a study partner, carefully
A chatbot makes a good quiz master if you give it the exam guide and make it show its sources. Upload the guide PDF, then paste this:
You are my study coach for the Databricks Certified Generative AI Engineer Associate exam. I have attached the official exam guide.
Rules:
- Only use objectives that appear in the attached guide.
- Use current Databricks product names. If the guide uses an older name (for example, Vector Search is now AI Search in the docs), give both.
- For every key claim, cite a docs.databricks.com page. If you can't verify something, say so instead of guessing.
Task: Quiz me on [SECTION NAME, e.g. Assembling and Deploying Apps] with 8 scenario-style multiple-choice questions, one at a time. Wait for my answer before showing the next question.
After each answer: say whether I was right, explain why in 3 sentences or fewer, and name the exam objective it tests.
At the end: list the objectives I missed and, for each, one hands-on task I can finish in Databricks Free Edition in under 30 minutes.If an answer arrives without a docs link, check it yourself. Our guide to making AI more accurate and honest has more ways to catch confident wrong answers.
Common mistakes
- Studying old names without translating them. A course that says “Vector Search” everywhere isn’t wrong, but you need to connect it to the current docs and to the guide’s wording.
- Trusting a practice-exam score. Databricks doesn’t offer official practice exams and says third-party results are often misleading. Use practice questions to find gaps, not to predict a pass.
- Skipping the small sections. Governance is only 8%, but four questions can decide a close result.
- Testing on a locked-down work laptop. Databricks warns that corporate firewalls can block the proctoring software and get your exam suspended. Run Kryterion’s system check days before.
- Forgetting the retake math. Fail, and you wait 14 days and pay the full fee again. Our Databricks exam cost breakdown covers retakes and recertification.
What to expect on exam day
- You register and reschedule through Webassessor, the exam platform from Databricks’ delivery partner, Kryterion.
- No notes or other test aids are allowed.
- Expect scenario questions, some asking you to choose two answers, like the samples in the guide.
- You see an unofficial result as soon as you finish. If you pass, Databricks reviews the proctor recording, then posts your badge to credentials.databricks.com and you get an email with a shareable badge.
Try it now
Open the exam guide, pick the heaviest section you know least about, and spend 10 minutes on step 1 of the hands-on checklist in Free Edition. Then test yourself with the free practice questions. Still weighing this against other credentials? Our ranking of AI certifications compares them side by side, and the AI exam prep hub has free practice for other popular AI certs.
🎁 Free download: The AI Cert Starter Kit
A one-page decision guide, a side-by-side comparison of the major AI certifications, and a 4-week study plan template, all in one free Notion doc. Free with your email address.
Related guides
- Databricks Generative AI Engineer Exam Cost
- Databricks Generative AI Engineer Practice Questions
- The Best AI Certifications in 2026, Ranked: Which Should You Get?
- AI Exam Prep
- AWS ML Engineer Associate (MLA-C01) Study Guide & Cheat Sheet
HOW TO // AI is not affiliated with or endorsed by Databricks. Exam details come from Databricks’ official certification pages and exam guide as of October 2026 and can change. Check the official exam page before you book.
Get an email when the Databricks Certified Generative AI Engineer Associate exam changes
We check the official exam pages every day. If the price, version, format or retirement date changes, you’ll get a short email. At most one a week, plus the HOW TO // AI newsletter on Thursdays.
Unsubscribe anytime. Privacy policy.
Frequently asked questions
How hard is the Databricks Generative AI Engineer Associate exam?
It's an associate-level exam, but it's hands-on in style. Databricks recommends 6+ months of experience with the tasks in the exam guide, and the questions are scenarios about building, deploying and evaluating RAG apps and agents. If you've built a small RAG app on Databricks yourself, it's very passable.
How long should I study for it?
Most people with some Python and LLM experience can prepare in about six weeks at 5 to 7 hours a week. If you already build RAG apps on Databricks, three or four weeks may be enough. If the platform is new to you, plan on eight.
Does Databricks publish the passing score?
No. Databricks says passing scores are set through statistical analysis and can change as exams are updated, so it doesn't publish them. You get an unofficial pass or fail result as soon as you finish.
Can I prepare without paying for a Databricks workspace?
Mostly, yes. Databricks Free Edition gives you a no-cost workspace for practice, with limits such as one AI Search endpoint and no Direct Vector Access indexes. Databricks also describes its Academy's on-demand courses as free.




