Free Databricks Certified Generative AI Engineer Associate Practice Questions

Sample Practice Questions

Try these 10 sample questions from the Databricks Certified Generative AI Engineer Associate practice bank. Each includes the answer and explanation. Continue with the interactive practice above to review the full question set.

1. 1A prompt produces inconsistent response layouts. Which addition most directly demonstrates the required format without changing the model?

   1. A. Several representative queries paired with ideal responses
   2. B. Several retrieved documents without response examples
   3. C. A longer list of synonyms for each incoming query
   4. D. A higher maximum token allowance for every response

   Answer: A

   Explanation: The guide says, "Include examples of well-formed queries and their corresponding ideal responses within the prompt template itself (few-shot learning)." These examples demonstrate the desired format; retrieval, synonyms, and a larger allowance do not demonstrate it.

   Reference: Improve RAG chain quality — Prompt augmentation — Provide examples

2. 2Which ordering prepares raw documents for a text-embedding index?

   1. A. Embed, parse, chunk, index
   2. B. Index, embed, parse, chunk
   3. C. Parse, clean, chunk, embed, index
   4. D. Chunk, index, parse, embed

   Answer: C

   Explanation: "Convert the chunked text data into a numerical vector representation" requires prepared text and chunks before embedding, followed by indexing.

   Reference: Build an unstructured data pipeline for RAG — Key components of the data pipeline

3. 3Which supported chat-model parameter should an engineer lower to reduce sampling randomness?

   1. A. max_tokens
   2. B. timeout
   3. C. temperature
   4. D. max_retries

   Answer: C

   Explanation: Temperature "Controls the randomness of the model’s output." The other parameters control length, waiting, or retries.

   Reference: Models — LangChain — Parameters — temperature

4. 4A custom PythonModel.predict must clean input x, call llm.invoke on the cleaned text, then parse the model response. Given those functions, which expression has the required order?

   1. A. clean(llm.invoke(parse(x)))
   2. B. llm.invoke(clean(parse(x)))
   3. C. parse(llm.invoke(clean(x)))
   4. D. llm.invoke(parse(clean(x)))

   Answer: C

   Explanation: predict "defines the model's prediction logic." Function composition evaluates clean(x), then llm.invoke, then parse. Each competing expression changes the stipulated preprocessing or post-processing order.

   Reference: MLflow PythonModel Guide — Define a custom PythonModel — predict

5. 5A sensitive number should remain partially visible, such as showing only its final digits. Which PIIMiddleware strategy matches this requirement?

   1. A. hash
   2. B. redact
   3. C. mask
   4. D. block

   Answer: C

   Explanation: mask means "Partially obscure". Redaction replaces the value, hashing substitutes a hash, and blocking raises an exception.

   Reference: Guardrails — LangChain — PII detection — Strategies — mask

6. 6Which listed judge requires reference ground truth to evaluate whether an answer is correct?

   1. A. Safety
   2. B. RelevanceToQuery
   3. C. RetrievalGroundedness
   4. D. Correctness

   Answer: D

   Explanation: Correctness compares responses to "the provided ground truth". Safety, query relevance, and grounding do not require reference answers in the documented table.

   Reference: Built-in LLM judges — Available judges — Correctness

7. 7A RAG response must interpret relative dates using today's date, which changes between requests. How should the prompt template receive that information?

   1. A. Fix the date in a few-shot example used for every request
   2. B. Inject a current-date variable when constructing each prompt
   3. C. Replace the retrieval query with a fixed calendar document
   4. D. Increase the response token limit before submitting each prompt

   Answer: B

   Explanation: The guide lists "variables such as current date, user context, or other relevant metadata." Supplying the changing date at inference time gives the template the required context. A fixed example becomes stale; replacing the query or changing token limits does not supply the current date.

   Reference: Improve RAG chain quality — Prompt augmentation — Parameterize your prompt template

8. 8Repeated navigation headers dominate retrieved website chunks. Which preprocessing change directly addresses this noise?

   1. A. Normalize whitespace while preserving the navigation text
   2. B. Convert navigation text to lowercase before chunking
   3. C. Normalize Unicode while preserving all navigation text
   4. D. Remove irrelevant navigation headers before embedding

   Answer: D

   Explanation: The guide recommends removing "irrelevant or noisy information, such as headers, footers, or special characters." Whitespace, case, and Unicode normalization leave unwanted navigation content in the corpus.

   Reference: Build an unstructured data pipeline for RAG — Best practices for parsing data — Data cleaning

9. 9A model's supported parameters include max_tokens. Which requirement does reducing it directly enforce?

   1. A. A larger input context window
   2. B. A lower network timeout
   3. C. A lower retry count
   4. D. A smaller response-token allowance

   Answer: D

   Explanation: max_tokens "Limits the total number of tokens in the response". It does not enlarge input capacity or configure networking.

   Reference: Models — LangChain — Parameters — max_tokens

10. 10Which PythonModel lifecycle method is intended to load additional context needed by a custom model?

    1. A. predict
    2. B. predict_stream
    3. C. log_model
    4. D. load_context

    Answer: D

    Explanation: "Implement the load_context method if the model requires additional context to be loaded." Prediction methods serve inference; log_model packages the model.

    Reference: MLflow PythonModel Guide — Define a custom PythonModel — load_context