You can see it againUnder pressure people bring back shapes and positions long after the wording has gone.
Picture superiority · Shepard 1967, Standing 1973
DATABRICKSDatabricks Certified Generative AI Engineer Associate
The Databricks Certified Generative AI Engineer Associate certification validates your ability to design, build, and deploy performant generative AI solutions on the Databricks platform. It is for engineers who develop LLM-enabled applications, including RAG systems and LLM chains, using Databricks tools like AI Search, Model Serving, MLflow, and Unity Catalog. Earning this credential demonstrates you can turn complex requirements into reliable, production-ready AI solutions.
Exam formatMultiple choice
Duration90 minutes
DeliveryDatabricks
Free questions644
Content last reviewed 30 July 2026 · Up to date
The certification
What Databricks Certified Generative AI Engineer Associate proves, and what it asks of you
What this certification covers, who it is written for, and what the exam itself looks like on the day.
6domains
32objectives
155concepts
$200exam fee
What it is
What this certification is
What it validates, who it is written for, and the experience it assumes.
About this certification
The Databricks Certified Generative AI Engineer Associate certification assesses your ability to design and implement LLM-enabled solutions using Databricks. This includes problem decomposition to break down complex requirements into manageable tasks, as well as choosing appropriate models, tools, and approaches from the current generative AI landscape for developing comprehensive solutions.
The exam also covers Databricks-specific tools such as AI Search for semantic similarity searches, Model Serving for deploying models and solutions, MLflow for managing the solution lifecycle, and Unity Catalog for data governance. Individuals who pass this exam can be expected to build and deploy performant RAG applications and LLM chains that take full advantage of Databricks and its toolset.
Who it’s for
This certification is for generative AI engineers and developers who design, build, and deploy LLM-enabled solutions on the Databricks platform. It is ideal for practitioners who work with RAG applications, LLM chains, and the Databricks toolset to solve real-world problems. You should be comfortable with Python, as all machine learning code in the exam is in Python, and have a working understanding of data manipulation concepts, which may be presented in SQL for non-ML workflows.
Recommended experience
6+ months of hands-on experience performing the generative AI solutions tasks outlined in the exam guide. Designing and implementing LLM-enabled solutions; Using Databricks tools such as AI Search, Model Serving, MLflow, and Unity Catalog; Building RAG applications and LLM chains; Problem decomposition and model selection
The syllabus
What you’ll learn
Every domain and objective Databricks measures, with the weight they carry on the exam.
The official Databricks exam outline · checked 30 July 2026 · See the source
Section 1: Design Applications
- Design a prompt that elicits a specifically formatted response
- Select model tasks to accomplish a given business requirement
- Select chain components for a desired model input and output
- Translate business use case goals into a description of the desired inputs and outputs for the AI pipeline
- Define and order tools that gather knowledge or take actions for multi-stage reasoning
- Determine how and when to use Agent Bricks (Knowledge Assistant, Multiagent Supervisor, Information Extraction) to solve problems
Section 2: Data Preparation
- Apply a chunking strategy for a given document structure and model constraints
- Filter extraneous content in source documents that degrades quality of a RAG application
- Choose the appropriate Python package to extract document content from provided source data and format.
- Define operations and sequence to write given chunked text into Delta Lake tables in Unity Catalog
- Identify needed source documents that provide necessary knowledge and quality for a given RAG application
- Use tools and metrics to evaluate retrieval performance
- Design retrieval systems using advanced chunking strategies
- Explain the role of re-ranking in the information retrieval process
Section 3: Application Development
- Model Selection and Evaluation
- Prompt Engineering and Augmentation
- Chunking and Embedding Strategies
- Safety and Guardrails
- Tooling and Frameworks
Section 4: Assembling and Deploying Applications
- Build and Configure RAG Applications
- Develop and Deploy LLM Applications
- Manage Model Lifecycle and Governance
- Integrate External Services and Interfaces
Section 5: Governance
- Use masking techniques as guard rails to meet a performance objective
- Select guardrail techniques to protect against malicious user inputs to a Gen AI application
- Use legal/licensing requirements for data sources to avoid legal risk
- Recommend an alternative for problematic text mitigation in a data source feeding a GenAI application
Section 6: Evaluation and Monitoring
- Evaluation Metrics and Model Selection
- Monitoring and Logging for Deployed Applications
- Agent Evaluation and Feedback
- Cost Control and Usage Tracking
- Evaluation Judges and Ground Truth
On the day
The exam itself
Everything Databricks publishes about sitting it, and nothing we inferred.
Prerequisites
No mandatory prerequisites — this certification has no required predecessor exam or credential.
CertificationDatabricks Certified Generative AI Engineer Associate
Exam formatMultiple choice
Duration90 minutes
Questions45 scored questions
DeliveryDatabricks
LanguagesEnglish, 日本語, Português BR, 한국어
Pricing$200
Certification levelAssociate
After you pass
Where this credential goes next
The path Databricks lays out, how the credential is kept, and where to book.
Step-by-step path to Databricks Certified Generative AI Engineer Associate
Credential earnedDatabricks Certified Generative AI Engineer Associate Associate level certification Renewal and maintenance
The Databricks Certified Generative AI Engineer Associate certification is valid for two years. To recertify, you must take the current version of the exam. Stay current with the latest technologies and maintain your certification.
Learn more about renewal requirements Lifecycle status
This certification is currently active and available. Databricks maintains this certification to validate current skills and industry relevance.
Exam status: ActiveMaintained by Databricks
Exam registration
Register for the exam through Databricks, Databricks’s authorized testing partner.
Schedule your examVisit the official Databricks certification page for exam policies and requirements.
View the official pageYour coach
And when you are serious, your coach Pip takes over
Your coach in the app reads what you have answered with the book closed and tells you one thing to do tonight. It will not count an answer you gave with the page open, and it will tell you when you are not ready.
See how the coach worksBefore you book
Questions people ask
Is there a hands-on or lab component in the exam?
The exam is proctored and consists of multiple-choice questions. There is no hands-on lab component; however, the exam may include unscored items for statistical purposes.
What happens if I fail the exam? Can I retake it?
Databricks does not publish a specific retake policy on the exam page. For detailed retake policies, refer to the Databricks Exam Terms and Conditions.
Are there any test aides allowed during the exam?
No test aides are allowed during the exam. You must rely on your own knowledge and experience.
How soon will I receive my exam results?
Databricks does not specify the exact timing for score reports on the exam page. Typically, results are available shortly after the exam, but you should check the exam delivery platform for details.
What job roles does this certification map to?
This certification is designed for generative AI engineers who design, build, and deploy LLM-enabled solutions on Databricks, including RAG applications and LLM chains.
Can I recertify by passing a different Databricks exam?
No. To recertify, you must take the current version of the Databricks Certified Generative AI Engineer Associate exam. Passing a different exam does not renew this certification.
Is the exam available in my region?
The exam is delivered online or at test centers. Online proctoring is available globally, while test center availability may vary by region. Check the exam delivery platform for options in your area.
Are there any discounts or vouchers available for the exam?
Databricks does not publish discount or voucher information on the exam page. Contact Databricks or check the exam delivery platform for any available offers.
Information freshness · Content last reviewed on 2026-07-30 Up to date
Practice free questions 644 questions, free, no account needed.
The curriculum
The whole exam, mapped and written
Every domain, every objective, and every concept Databricks measures — each one written out.
6domains
32objectives
155concepts
140free pages
The domains, and what sits inside them
Every objective below is a page you can open and practise now, without an account.
The official Databricks exam outline · checked 30 July 2026 · See the source
1Section 1: Design Applications120 questions
Design a prompt that elicits a specifically formatted response6 concepts · 4 pagesOpen →Select model tasks to accomplish a given business requirement4 concepts · 5 pagesOpen →Select chain components for a desired model input and output4 concepts · 4 pagesOpen →Translate business use case goals into a description of the desired inputs and outputs for the AI pipeline4 concepts · 5 pagesOpen →Define and order tools that gather knowledge or take actions for multi-stage reasoning5 concepts · 4 pagesOpen →Determine how and when to use Agent Bricks (Knowledge Assistant, Multiagent Supervisor, Information Extraction) to solve problems5 concepts · 4 pagesOpen →
2Section 2: Data Preparation158 questions
Apply a chunking strategy for a given document structure and model constraints5 concepts · 4 pagesOpen →Filter extraneous content in source documents that degrades quality of a RAG application4 concepts · 4 pagesOpen →Choose the appropriate Python package to extract document content from provided source data and format.3 concepts · 5 pagesOpen →Define operations and sequence to write given chunked text into Delta Lake tables in Unity Catalog5 concepts · 4 pagesOpen →Identify needed source documents that provide necessary knowledge and quality for a given RAG application4 concepts · 5 pagesOpen →Use tools and metrics to evaluate retrieval performance4 concepts · 4 pagesOpen →Design retrieval systems using advanced chunking strategies7 concepts · 5 pagesOpen →Explain the role of re-ranking in the information retrieval process5 concepts · 3 pagesOpen →
3Section 3: Application Development114 questions
4Section 4: Assembling and Deploying Applications74 questions
5Section 5: Governance70 questions
Use masking techniques as guard rails to meet a performance objective6 concepts · 3 pagesOpen →Select guardrail techniques to protect against malicious user inputs to a Gen AI application6 concepts · 4 pagesOpen →Use legal/licensing requirements for data sources to avoid legal risk3 concepts · 5 pagesOpen →Recommend an alternative for problematic text mitigation in a data source feeding a GenAI application3 concepts · 3 pagesOpen →
6Section 6: Evaluation and Monitoring108 questions
Evaluation Metrics and Model Selection3 concepts · 4 pagesOpen →Monitoring and Logging for Deployed Applications6 concepts · 5 pagesOpen →Agent Evaluation and Feedback6 concepts · 6 pagesOpen →Cost Control and Usage Tracking3 concepts · 5 pagesOpen →Evaluation Judges and Ground Truth4 concepts · 4 pagesOpen →
Not a table of contents
Every line on this map already opens
In front of every objective the practice pages are already there, free and without an account. This is one objective, opened.
Design a prompt that elicits a specifically formatted response4 pagesOpen →
Free in front of it
20 questions on this objective, five to a page. Every range above is a real page, open now, with no account.
Reading the map is not the same as knowing the ground
The curriculum tells you what is on the exam. Proving you know it is a different job — and it is the one the closed-book run does.
Free questions
644 questions. No account, no card.
The whole bank is open. 5 questions to a page, every answer explained, and a discussion thread on each one.
644questions
140pages
32objectives
Freealways
Browse by domain
Every objective, and every page range, is a link — so you can pick up exactly where you left off.
1Section 1: Design Applications120 questions
2Section 2: Data Preparation158 questions
3Section 3: Application Development114 questions
4Section 4: Assembling and Deploying Applications74 questions
5Section 5: Governance70 questions
6Section 6: Evaluation and Monitoring108 questions
What a practice page gives you
1Five to a page
Short enough to finish, long enough to matter.
2Every answer explained
Not only which one is right — why the others are wrong.
3A discussion on each
Ask, answer, and vote. Every question has its own thread.
A question, and the idea behind it
Every question comes from a concept
These are not trivia. Each one is written against a concept in the book, so when you get one wrong there is somewhere to go and find out why.
The visual book
See it once. Carry it into the exam.
The pages shown here come from our AI-900 book — an example of how each concept is written in plain language and, where the idea needs one, drawn as a full page you can take in at a glance.
155concepts
644free questions
Why a picture, and not another paragraph
Three reasons, and each one is a real finding rather than a slogan.
You can see it againThe whole idea at onceWhere it starts, what happens in the middle, what comes out, and the mistake to avoid.
Multimedia principle · MayerThe look-alikes sit togetherThe pairs the exam tests are drawn side by side, so the difference is seen, not told.
Dual coding · PaivioWhat one page is made of
You are never asked to read a poster here — only to see how one is built. After that, every other page is legible at a glance.
1A numbered strip on top
The idea as a sequence, followed with a finger before a word is read.
2Panels for the parts
What it is, how the machine learns it, when it is the right tool.
3The trap, framed
The distinction the exam tests, given its own box instead of buried in prose.
4One line at the bottom
The sentence to carry into the exam room.
Open book, on purpose
Answer it with the picture beside you
This is the part that teaches. The illustration and the written explanation stay where they are while you work, so a scenario stops being a memory test and becomes something you can simply look at.
It teaches · it does not count
A smartphone uses AI to unlock when the owner looks at the camera. Which AI capability is being used?
AImage classification
BFacial recognition
COptical character recognition
DObject detection
Look at the five steps. Detect finds a face, align straightens it, encode turns it into numbers, search compares it with a stored face, and decide answers one question: is this the same person? Classification would only say “a face”. Detection would only draw a box. Neither decides whose.
Then the page shuts, and it starts counting
The same questions come back with the book closed — that run is the one that counts. After it, your coach picks one thing for tonight, sized to the time you have, and brings pages back before you lose them.
Testing effect · Roediger & Karpicke 2006 · spacing effect · Cepeda et al. 2006
Official resources
Straight from Databricks
Where the exam is defined, scheduled and scored.
2official links
155concepts mapped
The source
Databricks’s own pages
We link to them rather than repeat them, so nothing here goes stale behind them.
Checked against the source
Every objective on the official Databricks outline already has questions here
We build from the official skills outline, not from a summary of it — 32 objectives, 155 concepts written under them, and free questions against every one. When Databricks changes the outline, this page changes with it.
32objectives on the outline→155concepts written→644free questions
An official page tells you what will be asked. It cannot tell you what you will get right.
That is the only question worth answering the night before, and no link answers it. You answer it by taking the questions with the book closed, and seeing what comes back.