CertifyDeckCargando contenido…
Databricks Generative AI Engineer Associate · Sample questions
10 real questions from the CertifyDeck Databricks Generative AI Engineer Associate bank, spread across the domains of the official Databricks Certified Generative AI Engineer Associate skills outline, with the correct answer and a full explanation of why the other options are wrong. They are original questions written from the public syllabus, not exam dumps, and they are the same style you will find in the practice tests and mock exams.
Correct answer: B. The engineer who creates the endpoint needs READ access to that secret.
For secrets-based environment variables, the requirement is that the endpoint creator has READ access to the Databricks secrets referenced in the configuration; the endpoint then fetches the secret itself at serving time. Callers only need permission to query the endpoint and never handle the secret, so granting them READ on the scope is unnecessary, and CAN MANAGE would give them far more than querying.
Correct answer: D. Call mlflow.genai.register_prompt() with the same prompt name and the corrected template
Prompt versions are immutable once created, so the only way to change a template is to register a new version; calling register_prompt() with an existing name creates the next version with its own commit message. Editing version 3 in place is not possible, and deleting an alias only removes the pointer while the version stays unchanged. Tags are key-value metadata and do not change the template that load_prompt() returns.
Correct answer: A. The provider's terms of use linked in the listing, describing the license grant and restrictions
Marketplace policy requires every listing to include the provider's terms of use describing the license grant, other pertinent terms and any restrictions, and products must comply with those terms. The USE MARKETPLACE ASSETS privilege and the Unity Catalog grants control who can access the data, not what the license permits. An instantly available product still requires accepting terms, so the label does not mean unrestricted use.
Correct answer: C. Retrieve context, augment the prompt with it, then generate the answer
A deterministic RAG chain retrieves the top results relevant to the request, augments a prompt by combining the request with that context, and then sends the augmented prompt to the LLM. Generating before augmenting means the model never sees the retrieved context, so the answer would not be grounded.
Correct answer: D. Pass them in custom_inputs and return the source links in custom_outputs
ResponsesAgent natively supports custom_inputs and custom_outputs for extra agent inputs, such as a session ID or client type, and for outputs, such as retrieval source links, that should not be included in the chat history. Appending the values to the messages would put them into the conversation history, which the requirement rules out.
Correct answer: C. Wrap the existing agent in the MLflow ResponsesAgent interface
Databricks recommends the MLflow ResponsesAgent interface, which lets you author agents with any third-party framework and wrap existing agents to integrate with Databricks logging, tracing, evaluation, deployment and monitoring. LangChain is not the only supported framework, converting every step into Unity Catalog functions is an unnecessary rewrite, and a Vector Search index stores and retrieves embeddings rather than hosting an agent.
Correct answer: C. The Review App, with its Chat UI and labeling sessions
The Review App gives domain experts a Chat UI to test a prototype and labeling sessions to record feedback and expectations, which MLflow stores on traces and can turn into evaluation datasets. The retrieval quality evaluation and the Safety judge are automated and involve no expert input, and querying an endpoint with curl requires technical skills and captures no structured feedback.
Correct answer: D. Expectations with expected_facts or an expected_response
Correctness requires ground truth: it takes inputs, outputs and expectations, and it reads the reserved expectations keys expected_facts or expected_response. The guidelines key is used by the Guidelines judges, not by Correctness. A retriever span is what the retrieval judges need, and outputs can be generated by MLflow from the trace during direct evaluation.
Correct answer: C. Work with domain experts and stakeholders on content and likely queries
The right corpus depends on the application's requirements, and Databricks recommends engaging domain experts and stakeholders from the start to identify and curate relevant content and the queries users are likely to submit. Ingesting everything brings in duplicates and irrelevant documents that degrade retrieval, embedding leaderboards rank models rather than documents, and a pre-trained LLM does not know proprietary content, which is why RAG is used.
Correct answer: D. RetrievalRelevance
RetrievalRelevance judges whether the retrieved context is relevant to the request and needs no ground truth. RetrievalSufficiency and Correctness both require expectations such as expected facts, and RelevanceToQuery evaluates the app's response rather than the retrieved documents.
Answer each question before opening the explanation, and write down why you discarded the other options. On Databricks Generative AI Engineer Associate, most mistakes come from choosing an option that is technically possible but does not meet every requirement in the scenario.
These 10 questions are a taste of the format, not a diagnostic. To know which domains you fail, practice by domain for free and take a timed mock exam that mirrors the official weights and duration.
Start for free, no sign-up or card. Every question includes an explanation like the ones above, and the mock exam reproduces the official Databricks Generative AI Engineer Associate format.