Databricks-GenAI-Assoc Practice Test 14 — 15 Questions

Question 1

A developer is creating a Databricks notebook to prototype a GenAI application. They need to install the `databricks-langchain` library to use LangChain integrations with Databricks. Which command should they use in the notebook?

Choose an answer to begin — your selection is scored in the full session.

Accepted Answer

%pip install databricks-langchain. In Databricks notebooks, the `%pip install` magic command is the recommended way to install Python libraries. It ensures the package is installed in the notebook's isolated environment and is immediately available for import. Using `%pip` with `databricks-langchain` correctly sets up the LangChain integrations needed for the GenAI application.

Answer

%conda install databricks-langchain

Answer

dbutils.library.installPyPI("databricks-langchain")

Answer

!pip install databricks-langchain

Question 2

A GenAI engineer is building a retrieval-augmented generation (RAG) application using Databricks Vector Search. They notice that the retriever sometimes returns irrelevant chunks that hurt answer quality. They want to add a reranking step to improve the relevance of retrieved documents before passing them to the LLM. Which component should they add to their RAG pipeline?

Accepted Answer

A cross-encoder reranker. A cross-encoder reranker is specifically designed to reorder retrieved documents by jointly encoding the query and each document, producing a relevance score. This step filters out irrelevant chunks and improves the quality of context passed to the LLM. In Databricks, you can incorporate a reranker as a separate component in a LangChain or custom RAG pipeline, often using a model served via Mosaic AI Model Serving or an external endpoint.

Answer

A larger embedding model

Answer

A fine-tuned LLM

Answer

A BM25 retriever

Question 3

A data engineer is preparing a large Delta table of conversation logs for embedding generation. The table has frequent small appends, and the engineer needs to reduce file fragmentation and improve read throughput before the embedding job runs. Which two actions should the engineer take? (Choose two.)

Accepted Answer

Enable auto compaction and optimized writes on the Delta table so small writes are coalesced automatically.. File fragmentation from frequent appends is addressed by consolidating small files. OPTIMIZE with optional ZORDER compacts and clusters existing data, while auto compaction and optimized writes prevent new fragmentation from accumulating. Together they reduce the number of files the embedding job must open and can improve predicate pushdown, directly improving read throughput.

Answer

Run VACUUM with a retention of zero hours to delete old files and free space before the embedding job.

Answer

Increase the number of shuffle partitions to the maximum supported value so that each task writes a single small file.

Answer

Convert the table to a Parquet directory and rely on the file system to merge small files during reads.

Question 4

A GenAI engineer has registered a fine-tuned Llama 3 model in Unity Catalog as `prod.ml.models.support_llm` and now needs to expose it as a REST endpoint for a customer-facing chatbot. The team wants Databricks to manage the serving infrastructure, GPU autoscaling, and version upgrades with minimal operational overhead. Which action should the engineer take?

Accepted Answer

Create a Mosaic AI Model Serving endpoint that serves the Unity Catalog model version.. Serving a Unity Catalog-registered model through Mosaic AI Model Serving is the standard Databricks pattern for production GenAI endpoints. It provides managed GPU compute, autoscaling, and versioned deployments without the engineer building infrastructure. The other choices either use batch or interactive compute that lacks an inference endpoint, or confuse storage with serving, so none can deliver a low-latency REST API for the chatbot.

Answer

Publish the model artifact to a Delta table and query it with a SQL warehouse.

Answer

Deploy the model to an all-purpose cluster and expose it through a notebook's REST API calls.

Answer

Register the model in the workspace model registry and call it from a Databricks job schedule.

Question 5

A GenAI engineer is building a multi-step agent that uses Databricks Foundation Model APIs. The agent must decide when to call a weather tool and when to answer directly. The engineer wants to ensure the agent's decision-making is reliable and that failures in tool calls are handled gracefully. Which design approach should the engineer use?

Accepted Answer

Implement a ReAct-style loop where the LLM outputs a thought, action, and action input, and the agent executes the tool and feeds back the observation.. A ReAct-style loop enables the agent to reason about whether a tool is needed, execute it, and observe the result before deciding the next step. This iterative process supports graceful error handling because exceptions can be captured and returned as observations, allowing the LLM to adjust. Single-pass or fixed-chain approaches lack this adaptability and feedback.

Answer

Chain multiple LLM calls where the first call always invokes the weather tool and the second call formats the answer.

Answer

Fine-tune the LLM on examples of weather queries and answers, then deploy it without tools.

Answer

Use a single prompt that instructs the LLM to output a JSON with either an answer or a tool call, and parse it once.

Question 6

A team is building a retrieval-augmented generation application on Databricks and wants to reduce hallucination by improving the quality of retrieved context before it reaches the LLM. Which TWO techniques should they apply? (Choose two.)

Accepted Answer

Apply a reranker model to the top-k retrieved chunks before passing them to the LLM.. Reranking retrieved candidates with a cross-encoder and improving chunking with overlap and metadata both raise the relevance of the context supplied to the LLM, which directly reduces hallucination. Raising temperature, stuffing the full corpus, and dropping semantic search all degrade grounding rather than improve it.

Answer

Increase the LLM temperature to encourage more diverse answers.

Answer

Disable the Vector Search index and rely on keyword matching only.

Answer

Store the entire document corpus in the prompt for every request.

Question 7

A team is preparing a large text corpus for embedding generation with a foundation model endpoint on Databricks. They must reduce token cost and improve retrieval quality before vectorization. Which two preprocessing steps should be applied to the raw text? (Choose two.)

Accepted Answer

Deduplicate near-identical passages and drop documents that fail a minimum content-length threshold.. Removing boilerplate and normalizing whitespace cuts wasted tokens while sharpening the text signal, and deduplicating plus dropping undersized fragments avoids paying to embed redundant or low-information content. Uppercasing, stop-word stripping, and blind truncation either damage semantics or discard content, so they do not improve cost or retrieval quality.

Answer

Remove stop words and punctuation from every passage before generating embeddings.

Answer

Truncate every document to its first 512 characters to guarantee uniform chunk size.

Answer

Convert all text to uppercase before tokenization to standardize casing.

Question 8

A GenAI engineer is deploying a chain that calls an external LLM API. The chain must not block the serving thread while waiting on the remote API, and the endpoint must handle many concurrent requests. Which implementation approach should the engineer choose when logging the model?

Accepted Answer

Implement predict as an async function that awaits an async HTTP client, so the event loop can interleave other requests while waiting on the external API.. Calling an external LLM API is I/O-bound, so the serving process should not block a thread while waiting. Writing predict as an async function with an async HTTP client lets the event loop serve other requests during those waits, improving concurrency on each replica. Synchronous clients, table-based queues, and single-concurrency settings all undermine throughput or break the synchronous API contract.

Answer

Batch all incoming requests by writing them to a Delta table and have a separate job poll the table and call the external API.

Answer

Use a synchronous HTTP client inside predict and increase the endpoint's scale-to-zero timeout so queued requests eventually complete.

Answer

Set the endpoint's concurrency to one and rely on the external API's own load balancing to absorb parallel requests.

Question 9

A data engineer is preparing a large corpus of support tickets stored in a Unity Catalog volume for fine-tuning a Llama model on Databricks. They must remove personally identifiable information (PII) before the data reaches the training cluster. Which TWO approaches are appropriate for detecting and redacting PII at scale in this pipeline? (Choose two.)

Accepted Answer

Apply a Spark NLP or presidio-based pandas UDF that detects entities such as names, emails, and phone numbers and replaces them with placeholders.. PII must be removed from the content itself before training. Distributed detectors such as Presidio or Spark NLP in pandas UDFs, and model-based rewriting with ai_query(), both transform the text at scale inside Databricks. Column masks, encryption, and retention policies govern access or lifecycle but leave the underlying tokens intact, so they do not satisfy the preprocessing requirement.

Answer

Set the table's retention policy to 0 days so that raw tickets are deleted immediately after ingestion.

Answer

Use Unity Catalog column masks on the raw text column so that anyone querying the table sees redacted values.

Answer

Enable server-side encryption on the Unity Catalog volume and rely on the storage layer to anonymize the text.

Question 10

A GenAI engineer is building a customer-support assistant on Databricks. The assistant must invoke an external LLM endpoint that requires an API token, and the token must not appear in notebook source or logs. The engineer wants to store the token once and let notebooks and Databricks Jobs reference it by name. Which Databricks capability should the engineer use to satisfy this requirement?

Accepted Answer

Databricks secrets backed by a secret scope. Databricks secrets provide a governed, auditable store for sensitive values and expose them through dbutils.secrets.get or secret references, so the external LLM token is never embedded in code or logs. Other storage locations either expose plaintext to readers or serve a different purpose, such as authenticating to the workspace rather than to an external provider.

Answer

A Unity Catalog volume mounted at /Volumes/secure/token.txt

Answer

A workspace environment variable set in the cluster Spark configuration

Answer

A personal access token stored in the Databricks CLI configuration file

Question 11

A data science team has registered a new version of a retrieval-augmented generation (RAG) agent in Unity Catalog as a model named main.genai.support_agent. They need to make this version available as a low-latency REST endpoint with automatic scaling and no cluster management. Which Databricks capability should they use?

Accepted Answer

Mosaic AI Model Serving, by creating a serving endpoint that serves the registered Unity Catalog model.. Deploying a registered Unity Catalog agent model as a Mosaic AI Model Serving endpoint gives the team a managed REST inference API with automatic scaling and no cluster administration. It preserves Unity Catalog governance and lineage, supports low-latency online serving, and is the intended deployment path for GenAI agents on Databricks. Batch, interactive, and SQL compute cannot provide the same managed serving contract.

Answer

Databricks SQL warehouses, by registering the agent as a SQL function and querying it from a dashboard.

Answer

All-purpose interactive clusters, by starting a cluster and calling the model through a notebook.

Answer

Jobs compute, by scheduling the agent as a nightly batch task and writing predictions to a Delta table.

Question 12

A team deploys a RAG agent to a Mosaic AI Model Serving endpoint. During load testing, requests intermittently return HTTP 429 responses even though the endpoint shows healthy replicas. The agent calls an external vector search index and a foundation model endpoint. Which action most directly addresses the 429 responses?

Accepted Answer

Increase the endpoint's provisioned concurrency and configure appropriate rate limits so the agent can handle the incoming request rate.. A 429 from a Mosaic AI Model Serving endpoint signals that inbound request rate exceeds the endpoint's provisioned capacity or configured rate limits. Raising provisioned concurrency and setting realistic rate limits aligns capacity with the load-test traffic and directly removes the rate-limiting responses. Client retries, smaller prompts, and index relocation do not increase the endpoint's ability to accept concurrent requests, so they leave the root cause unaddressed.

Answer

Move the vector search index into the same Unity Catalog schema as the agent model to reduce cross-service latency.

Answer

Add retry logic with exponential backoff and jitter to the client calling the endpoint.

Answer

Switch the agent's foundation model calls to a smaller context window to reduce token usage per request.

Question 13

A GenAI engineer is building a retrieval-augmented chatbot whose answers must cite the exact source document and page. The team wants the chatbot's responses to include structured citations that downstream UIs can render. Which Databricks feature should the engineer use to return these structured citations from the model endpoint?

Accepted Answer

Mosaic AI Agent Framework's document citation support, where the retriever returns chunks with metadata and the agent formats them into a structured citations field in the response.. Structured citations must be part of the model endpoint's response contract, not a side channel. The Mosaic AI Agent Framework's citation support lets the retriever supply chunk metadata and the agent emit a citations field the UI can render directly. Observability traces, retrieval reranking, and payload logging serve different purposes and cannot substitute for returning structured citations to the caller at inference time.

Answer

Databricks Vector Search's built-in reranking, which automatically appends source metadata to the model's generated answer.

Answer

MLflow Tracing with span attributes that capture the retrieved document IDs and page numbers for each LLM call.

Answer

Mosaic AI Model Serving's payload logging, which writes each request and response to an inference table that the UI can read for citations.

Question 14

A GenAI engineer is developing a conversational agent using Databricks. The agent must maintain context across multiple turns and retrieve relevant information from a knowledge base. They want to ensure the agent can handle follow-up questions that refer to previous exchanges. Which TWO techniques should they implement to manage conversation state and retrieval effectively? (Choose two.)

Accepted Answer

Store the conversation history in a Delta table and include it in the prompt for each turn.. Storing conversation history in a Delta table ensures persistence and allows inclusion of past turns in the prompt for context. Rewriting follow-up questions into standalone queries using that history improves retrieval accuracy by making the query self-contained. Together, these techniques enable the agent to handle multi-turn conversations effectively, where follow-up questions depend on earlier exchanges, and ensure the retrieval step has the necessary context.

Answer

Cache all responses from the LLM to avoid re-processing similar queries.

Answer

Use a separate Vector Search index for each conversation to retrieve relevant past messages.

Answer

Use a fixed window of the last three messages as context, ignoring older messages.

Question 15

A Generative AI engineer is packaging a retrieval-augmented generation application for deployment with Databricks Asset Bundles. The bundle must provision a Databricks job that periodically refreshes a Delta table used as the vector index, and the job requires a specific cluster node type that differs between the development and production workspaces. Which approach correctly handles the node type difference while keeping a single bundle definition?

Accepted Answer

Define the node type in a databricks.yml variable and override its value in a target-specific block for each environment.. The bundle should declare a variable for the node type and override it inside each target, because Databricks Asset Bundles resolve variables per target at deploy time. This keeps one authoritative job definition while letting development and production use different compute, and it avoids duplicated YAML that drifts over time.

Answer

Create a separate bundle directory for each environment with entirely duplicated resource files and deploy each one independently.

Answer

Use a notebook widget inside the job to read the node type from a table at runtime and restart the cluster with the new value.

Answer

Hard-code the production node type in the job resource and rely on the development workspace to silently substitute a smaller node type.