Practice Exam: Certified Generative AI Engineer Associate

Question bank: 105 · timed mock exam: 90 min

Official exam page on the vendor's site (opens in a new tab) Official exam guide (opens in a new tab) Official prep course (opens in a new tab)

Exam guide: format, domains and how to prepare

Compare: NCA-GENL vs Databricks Generative AI Engineer Associate

  • Application Development 29 questions
  • Assembling and Deploying Applications 25 questions
  • Evaluation and Monitoring 16 questions
  • Data Preparation 15 questions
  • Design Applications 12 questions
  • Governance 8 questions
Study mode (free) Exam mode (subscribers)

Free: 25 questions in study mode, all domains, no account.

Subscription: the 105 questions of this certification, the timed exam mode, and all other certifications too.

Subscribe · €9.99 / month

Secure card payment (Stripe) · No commitment · Cancel online

How to prepare for Certified Generative AI Engineer Associate with InfraPrep

The 105 questions cover 6 domains: Application Development, Assembling and Deploying Applications, Evaluation and Monitoring, Data Preparation, Design Applications, Governance. 25 of them are free, without an account.

Study mode

Answer at your pace: each answer is corrected straight away, with an explanation and the link to the documentation it relies on.

Exam mode

A timed exam of 90 minutes, with a question map to flag and come back to questions before you submit.

Results by domain

Your score is broken down by domain, so you know what to review first.

Original questions

Every question is written from the public documentation, never copied from the exam. You can contest an answer from study mode.

10 free sample questions

Question 1 Design Applications

A Generative AI Engineer is building a pipeline that sends product reviews to a chat model served by Foundation Model APIs. A downstream job parses every reply and fails whenever a field is missing or named differently. The team may later switch to a model from another provider and does not want to rewrite the request. What should the engineer do so that each reply follows a fixed set of JSON fields?

  1. Lower the temperature to 0 so that every reply uses the same field names
  2. Set max_tokens to a small value so the model cannot add extra fields
  3. Pass a JSON schema in the response_format field of the chat request
  4. Enable inference tables so malformed replies are logged and retried
Show the answer and the explanation

Correct answer: C

Databricks structured outputs are requested through the response_format field, which can carry a JSON schema, and the same request format works regardless of the model provider.

Source: official documentation

Question 2 Data Preparation

A company's internal runbooks are written in Markdown. Each runbook has clear headings (## Symptoms, ## Cause, ## Fix), and each section is self-contained. A first prototype split the text every 1,000 characters, and retrieved chunks often start in the middle of one section and end in the next. Which chunking strategy should the engineer try next?

  1. Keep fixed-size splitting but double the chunk size to 2,000 characters
  2. Split on the Markdown headers so each chunk follows a section boundary
  3. Remove all overlap between chunks so sections no longer bleed together
  4. Embed each full runbook as a single chunk regardless of its length
Show the answer and the explanation

Correct answer: B

Markdown has an inherent structure that can define chunk boundaries, so format-specific chunking on headers keeps each self-contained section together.

Source: official documentation

Question 3 Application Development

An engineer is prototyping a LangGraph agent in a notebook and wants to give it a retriever tool that queries an existing Databricks AI Search index, then bind that tool to the LLM to test tool calling locally. Which package does Databricks point to for building this retriever tool?

  1. databricks-langchain, part of Databricks AI Bridge
  2. mlflow-skinny, the lightweight MLflow client
  3. databricks-connect, the remote Spark client
  4. langchain-text-splitters, the chunking library
Show the answer and the explanation

Correct answer: A

Databricks AI Bridge packages such as databricks-langchain (and databricks-openai) provide helpers such as VectorSearchRetrieverTool to build an AI Search retriever tool locally.

Source: official documentation

Question 4 Design Applications

An insurance analytics team stores free-text claim notes in a Unity Catalog table. The business wants each note tagged with one of five fixed categories (theft, water damage, fire, liability, other) directly in a SQL query, with as little custom prompt work as possible. Which approach best matches this requirement?

  1. Call ai_extract with a schema listing the five categories as fields
  2. Call ai_parse_document on the notes to obtain their layout first
  3. Call ai_query with a hand-written prompt listing the five categories
  4. Call ai_classify with the five categories supplied as the labels
Show the answer and the explanation

Correct answer: D

The task is classification into a fixed label set, which is exactly what the task-specific ai_classify function does, and Databricks recommends starting with a task-specific AI function when one matches the objective.

Source: official documentation

Question 5 Data Preparation

A RAG assistant is built on 4,000 product manuals exported as PDFs. Every page carries the same header with the company address and a footer with a copyright line. Reviewers notice that many retrieved chunks are dominated by this repeated text. Which pipeline change addresses the root cause?

  1. Clean the parsed text to remove the repeated headers and footers
  2. Increase the number of retrieved chunks from 5 to 20 per query
  3. Switch the embedding model to one with a larger embedding window
  4. Add the copyright text to the system prompt as an instruction
Show the answer and the explanation

Correct answer: A

The noise comes from the source documents. Cleaning the extracted text to remove irrelevant content such as headers and footers reduces the unnecessary information the RAG chain must process.

Source: official documentation

Question 6 Application Development

While reviewing responses from a RAG assistant, a reviewer finds answers that state warranty durations that do not appear anywhere in the chunks the retriever returned. The team has no ground-truth answers yet. Which built-in MLflow LLM judge targets this issue most directly?

  1. Correctness
  2. RetrievalGroundedness
  3. RetrievalSufficiency
  4. ToolCallCorrectness
Show the answer and the explanation

Correct answer: B

The issue is a response that is not supported by the retrieved context (a hallucination). RetrievalGroundedness evaluates whether the response is grounded in the provided context, and it does not require ground truth.

Source: official documentation

Question 7 Design Applications

A RAG chatbot answers the first question of a conversation well, but follow-up questions such as "and what about the second option?" retrieve unrelated chunks, because the raw follow-up text is sent directly to the retriever. Which chain component should the engineer add, and where?

  1. A query preprocessing step that rewrites the request before retrieval
  2. A post-processing step that adds citations after generation
  3. A larger top-k value in the retrieval step with no other change
  4. A content moderation filter applied to the generated response
Show the answer and the explanation

Correct answer: A

The retrieval query is the problem. A preprocessing step before retrieval can use a model to rewrite the follow-up into a standalone query (for example by paraphrasing the conversation history) that better represents the user's intent.

Source: official documentation

Question 8 Data Preparation

The knowledge source for a new RAG application is a set of saved web pages (.html files) from an internal wiki. The engineer needs to navigate the page structure and extract the article text before chunking. Which Python package is designed for this parsing step?

  1. PyPDF2
  2. BeautifulSoup
  3. Spark ML MinHashLSH
  4. LangChain CharacterTextSplitter
Show the answer and the explanation

Correct answer: B

BeautifulSoup (like lxml) is an HTML parsing library that navigates the HTML structure and extracts the desired text from web pages.

Source: official documentation

Question 9 Application Development

A team is choosing between two chunking strategies for its RAG application. With strategy 2, the final answers score slightly better on correctness, but the team cannot tell whether the retriever actually improved or the LLM simply compensated. How should the team evaluate the two strategies before choosing one?

  1. Compare only final answer correctness, since that is what users see
  2. Compare only total token count, since smaller chunks cost less
  3. Measure retrieval metrics and response metrics for each strategy
  4. Pick the strategy with the smaller chunks, since they are more precise
Show the answer and the explanation

Correct answer: C

A RAG application can answer poorly with correct context, or well despite faulty retrieval. Measuring both retrieval metrics (such as chunk relevance or document recall) and response metrics is the only way to see what the chunking change actually did.

Source: official documentation

Question 10 Design Applications

A procurement manager says: "We receive thousands of supplier invoices as PDFs and want a table with the vendor name, invoice total, and due date for each one." When translating this goal into the inputs and outputs of an AI pipeline built with Agent Bricks Information Extraction, which description is correct?

  1. Input: a Delta table of totals; output: a free-form summary of each vendor
  2. Input: unstructured invoice documents; output: fields that follow a defined schema
  3. Input: a user question in chat; output: an answer with document citations
  4. Input: precomputed embeddings; output: the most similar invoice documents
Show the answer and the explanation

Correct answer: B

Information Extraction takes unstructured documents or text (such as PDFs) and produces structured insights that follow a schema you define, which matches a table of vendor, total, and due date.

Source: official documentation

Frequently asked questions

Are there free questions, and do I need an account?

Yes. 25 questions of this certification are free in study mode, across all its domains, without a subscription and without registration. The account is created when you subscribe, at checkout.

How does the subscription work, and can I cancel?

One monthly subscription covers all certifications and the timed practice exams. There is no minimum period. You cancel online, at any time, with the "Cancel your contract" function in My account; cancellation takes effect at the end of the month already paid, and no further payment is taken.

Where do the questions come from?

Every question is original, written from the vendor's public documentation. Each answer comes with an explanation and a link to the documentation page it relies on. InfraPrep is not affiliated with the vendors, and certification names belong to their owners.

In which language are the questions?

The questions are written in English, also on the French version of the site.

Is the catalog updated?

Yes, the catalog changes over time: new certifications and new questions are added.