Databricks-GenAI-Assoc Application Development — 30 Questions

Question 1

A team is developing a custom application using the Databricks SQL Warehouse. They need to optimize query performance for high-frequency small requests. Which TWO strategies should they implement?

Accepted Answer

Use Serverless SQL Warehouses to minimize cold-start latency.. For high-frequency small requests, minimizing overhead is key. Serverless SQL Warehouses provide faster startup times, reducing latency for bursty workloads. Additionally, using specific query syntax to avoid full table scans helps keep the execution plan lean. These strategies are fundamental for developers aiming to build responsive applications that rely on Databricks as the backend data engine, ensuring cost efficiency and optimal user experience in production environments.

Answer

Always set the warehouse size to '4XL' to ensure maximum compute power.

Answer

Disable the query result cache to ensure real-time data retrieval.

Answer

Store all application data in JSON files to optimize read performance.

Question 2

A GenAI application requires low-latency retrieval of vector embeddings. Which TWO components should the developer prioritize to minimize latency?

Accepted Answer

Implement Mosaic AI Vector Search to avoid manual index maintenance.. To minimize retrieval latency, developers must optimize the storage layer and the indexing strategy. Using vector-optimized search indexes within the Databricks environment and keeping data co-located with the compute resources significantly reduces I/O overhead. Furthermore, ensuring that the model serving infrastructure is provisioned with adequate capacity prevents bottlenecking. These architectural choices are vital for building responsive AI applications that meet the strict performance requirements of real-time user-facing services.

Answer

Store all vectors in a compressed Parquet file on S3 for batch retrieval.

Answer

Provision dedicated, high-memory clusters solely for the embedding model.

Answer

Increase the vector dimension size to ensure higher search accuracy.

Question 3

Which THREE steps are necessary to ensure a Databricks notebook can be executed as a production job with secure access to data in Unity Catalog?

Accepted Answer

Use a Service Principal to execute the job.. Production jobs require identity management and proper permissioning. By using a Service Principal, you ensure that the job runs under a dedicated, non-user-bound identity, which is essential for stable production environments. Coupling this with proper Grant statements in Unity Catalog ensures the job has the principle of least privilege. Finally, using cluster-level access policies prevents unauthorized configuration changes, creating a secure, repeatable, and compliant production execution environment.

Answer

Assign the job to the 'admin' user to avoid permission errors.

Answer

Store the job execution logs in a public cloud bucket.

Question 4

A developer is building a RAG application using Mosaic AI Model Serving. They need to ensure the application remains highly available during model updates. Which deployment strategy should be implemented to minimize downtime while testing the new model version?

Accepted Answer

Implement a blue-green deployment strategy by creating a new endpoint.. Blue-green deployment allows teams to run two versions of the serving endpoint simultaneously, ensuring seamless transition. By shifting traffic incrementally, developers can validate performance metrics against the production baseline without affecting end-user requests. This architectural pattern is critical in Databricks for maintaining SLAs while integrating iterative model improvements into live GenAI applications, effectively reducing the risk of downtime or performance degradation during model swaps.

Answer

Directly update the existing endpoint configuration with the new model version.

Answer

Pause the model serving endpoint before deploying the new version.

Answer

Scale the model serving endpoint to zero and redeploy.

Question 5

A developer is using MLflow to track a fine-tuning job. After the job completes, they notice the model weights are not appearing in the Unity Catalog. Which requirement must be met to ensure the model is governed and discoverable?

Accepted Answer

Set the registry URI to 'databricks-uc' and use a three-level namespace.. To register models in Unity Catalog, the model must be logged with the MLflow flavors compatible with the Databricks Model Registry. Specifically, it must be logged to the Unity Catalog-enabled registry rather than the workspace-local registry. This integration is vital for centralized lineage, permission management, and lifecycle tracking, ensuring that production-ready LLMs are fully compliant with enterprise governance standards before they are consumed by downstream inference applications.

Answer

Set the MLflow registry URI to the workspace-local path.

Answer

Convert the model to a generic pickle format before logging.

Answer

Manually upload the model weights to the DBFS root.

Question 6

A developer needs to optimize the latency of a RAG application. Which TWO techniques should be applied to improve retrieval performance from a Vector Search index?

Accepted Answer

Utilize approximate nearest neighbor (ANN) search algorithms.. Performance optimization in RAG requires both efficient retrieval and effective data management. By implementing approximate nearest neighbor (ANN) search and pruning irrelevant documents from the index, the system reduces the search space significantly. These techniques are crucial for maintaining low-latency responses, as they minimize the computational overhead of calculating distances in high-dimensional vector spaces, ultimately leading to faster and more relevant context retrieval for the generative model during inference.

Answer

Increase the number of dimensions in the vector embeddings.

Answer

Perform exhaustive exact search to ensure 100% precision.

Answer

Cache all user queries in a global memory store.

Question 7

A data engineering team is building a Databricks Asset Bundle (DAB) to automate the deployment of a RAG-based LLM application across development, staging, and production workspaces. They need to parameterize the LLM endpoint name based on the target deployment environment. How should the team configure the DAB configuration file?

Accepted Answer

Define target-specific blocks in databricks.yml using bundle environment variables and reference them via interpolation syntax in resource definitions.. Databricks Asset Bundles utilize YAML configuration files supporting environment-specific overrides and variable substitution using interpolation syntax like ${bundle.environment}. This enables developers to maintain a single codebase while dynamically injecting environment-appropriate resource names, such as different Model Serving endpoint names, ensuring seamless promotion across development, staging, and production environments without manual code modifications.

Answer

Hardcode the production endpoint name directly in the Python application script and use environment branching logic.

Answer

Create three entirely separate repository checkouts, each containing manually edited configuration files for a single environment.

Answer

Store the endpoint name in a temporary SQLite database on the driver node during job execution initialization.

Question 8

An AI application developer is configuring a Databricks Model Serving endpoint for a high-volume LLM application. Which TWO configurations or practices should the developer implement to optimize throughput and manage costs effectively? (Choose TWO)

Accepted Answer

Configure autoscaling with minimum and maximum scale limits to automatically adjust compute resources based on incoming traffic volume.. Optimizing Databricks Model Serving endpoints requires configuring appropriate autoscaling policies to handle traffic fluctuations without over-provisioning expensive GPU resources. Additionally, setting up endpoint tags, monitoring metrics via system tables, and enabling provisioned throughput options ensure cost predictability and high availability under heavy production workloads.

Answer

Provision the maximum possible GPU instance type permanently without enabling autoscaling to eliminate scaling latency.

Answer

Disable all rate limiting and concurrent request constraints to allow unlimited traffic processing on a single node.

Question 9

An MLflow developer is logging a custom PyFunc model that performs text summarization using a Hugging Face pipeline. When evaluating the model locally, the developer notices that the artifact fails to load in a fresh Python session due to missing dependency tracking. What is the best practice to ensure MLflow correctly captures all required packages?

Accepted Answer

Pass a dictionary or file defining the exact pip requirements via the extra_pip_requirements parameter during mlflow.pyfunc.log_model.. When logging models with MLflow, explicitly providing the conda_env or extra_pip_requirements parameter ensures that all required Python libraries, including specific framework versions, are serialized into the model metadata. This guarantees that when the model is loaded in Databricks Model Serving or downstream scoring jobs, the execution environment reconstructs successfully without missing dependencies.

Answer

Rely on global environment variables configured on the cluster driver node to inject libraries dynamically at inference time.

Answer

Manually copy the site-packages folder from the development laptop into the root of the MLflow artifact directory.

Question 10

Which Databricks feature is essential for maintaining the lineage of an AI model from the raw data used for training to the final deployed endpoint?

Accepted Answer

Unity Catalog. Unity Catalog acts as the central governance layer for all data and AI assets in Databricks. It automatically captures lineage between tables, models, and notebooks. This is essential for AI applications, as it allows organizations to trace model performance and bias back to specific training datasets, ensuring compliance and transparency. Without this integrated lineage, debugging model behavior and validating data provenance becomes an impossible task in complex, production-grade AI systems.

Answer

Databricks Delta Sharing

Answer

Databricks SQL Dashboards

Answer

Databricks Repos

Question 11

An organization is migrating their LLM application to Databricks. They require a centralized location to manage model versions, staging environments, and deployment status. Which feature acts as the primary registry for these requirements in the Databricks ecosystem?

Accepted Answer

Unity Catalog. Unity Catalog acts as the centralized model registry in Databricks, providing a unified interface for managing model lifecycle stages (e.g., Staging, Production, Archived). By registering models in Unity Catalog, teams gain granular access control, centralized lineage, and seamless integration with other platform services, ensuring that model versions are governed, searchable, and easily deployable across different environments within the workspace or across the entire organization.

Answer

Delta Lake

Answer

MLflow Tracking Server

Answer

Databricks Workflows

Question 12

A GenAI engineer is optimizing a RAG application for better retrieval accuracy. They decide to use Vector Search to improve the context provided to the LLM. Which THREE steps are mandatory to set up an effective Databricks Vector Search index? (Select THREE)

Accepted Answer

Create a Delta table containing the text chunks and metadata.. To set up an effective Vector Search index, the engineer must first create a Delta table containing the source text chunks. Next, they must define an endpoint to host the index. Finally, they must create the Vector Search index itself, which involves configuring the embedding model and specifying the source table. These components work together to index the data and allow for efficient semantic similarity searches within the Databricks environment.

Answer

Enable Unity Catalog on the workspace.

Answer

Install an external vector database like Pinecone or Milvus.

Question 13

Which service should an engineer use to programmatically manage the lifecycle of a Databricks Model Serving endpoint, including creating, updating, and deleting endpoints via API calls?

Accepted Answer

Databricks Model Serving API. The Databricks Model Serving API is the correct service for programmatically managing endpoint lifecycles. It provides REST endpoints that allow developers to automate model deployments within CI/CD pipelines. By integrating these API calls into workflow tools, engineers can ensure consistent, version-controlled deployments of their LLM applications, reducing the risk of manual configuration errors and accelerating the time-to-market for new iterations of their machine learning models.

Answer

Databricks SQL API

Answer

Unity Catalog API

Answer

Jobs API

Question 14

A GenAI engineer is building a retrieval-augmented generation (RAG) application using Databricks Vector Search. They need to ensure that the application can retrieve relevant documents even when user queries use synonyms not present in the source documents. Which approach should they take to improve retrieval recall while keeping the index up to date?

Accepted Answer

Use a dense embedding model that captures semantic similarity, and configure the Vector Search index to automatically sync with the source Delta table.. Using a dense embedding model that captures semantic similarity ensures that synonyms are represented closely in vector space, directly improving recall for synonym queries. Configuring automatic sync with the source Delta table keeps the index updated as documents are added or modified, which is essential for a production RAG application.

Answer

Create a separate keyword index using Apache Lucene and query both indexes, merging results with reciprocal rank fusion.

Answer

Increase the number of nearest neighbors returned by the Vector Search query to 100 and filter results by a minimum similarity score.

Answer

Enable hybrid search on the Vector Search index, which combines dense vector similarity with keyword-based BM25 scoring.

Question 15

A developer is using MLflow to track experiments for a RAG application. They want to log the retrieval component, which includes a Databricks Vector Search index and a reranker, as a single MLflow model. They also want to ensure that the model can be deployed to Mosaic AI Model Serving and automatically update the index when new documents are added. Which MLflow model flavor should they use to package the retrieval component?

Accepted Answer

mlflow.pyfunc. The pyfunc flavor is the most versatile MLflow model flavor, allowing you to package any Python code as a model. For a RAG retrieval component that integrates Databricks Vector Search and a reranker, pyfunc enables you to define custom inference logic, manage dependencies, and deploy seamlessly to Mosaic AI Model Serving. It also supports automatic index updates if the code includes logic to refresh the index, though that would typically be handled outside the model serving endpoint.

Answer

mlflow.transformers

Answer

mlflow.langchain

Answer

mlflow.sklearn

Question 16

A developer is building a RAG application using Databricks Mosaic AI Model Serving. They need to log the model with MLflow and deploy it to a serving endpoint. The model must accept a JSON payload with a 'query' field and return a response from an LLM, while also retrieving relevant documents from a Vector Search index. Which approach correctly implements this?

Accepted Answer

Log a custom PyFunc model that uses the Vector Search client to retrieve documents, then calls the LLM endpoint, and deploy it with MLflow.. The correct approach is to use a custom PyFunc model because it gives full control over the RAG workflow. You can integrate the Databricks Vector Search client to fetch relevant documents and then call an LLM endpoint. MLflow logs the model with its dependencies, and Mosaic AI Model Serving deploys it as a scalable endpoint. This pattern is flexible and production-ready.

Answer

Use MLflow's built-in langchain flavor to log a chain that includes Vector Search retrieval and LLM calls, then deploy to a serving endpoint.

Answer

Create a Databricks Job that runs a Python script to retrieve documents and call the LLM, and expose it via a REST API using Flask.

Answer

Log the LLM using mlflow.transformers and deploy it directly, relying on the model to retrieve documents internally.

Question 17

An AI engineer is deploying a RAG application on Databricks Model Serving. The application uses a Vector Search index to retrieve context and a foundation model endpoint to generate answers. During load testing, they observe that end-to-end latency is high, with Vector Search queries taking up to 2 seconds. Which configuration change is most likely to reduce latency while maintaining retrieval quality?

Accepted Answer

Switch from a direct vector index to a managed vector index with optimized embedding storage and quantization.. Managed vector indexes in Databricks Vector Search are engineered for low-latency queries, with features like quantization and optimized storage. Switching from a direct index to a managed index can reduce query latency without sacrificing retrieval quality, making it the most effective change.

Answer

Reduce the embedding dimension by using a smaller model, such as a 384-dimensional model instead of a 768-dimensional one.

Answer

Enable the Vector Search index's automatic sync and increase the sync frequency to every minute.

Answer

Increase the number of nearest neighbors returned by the Vector Search query from 5 to 20.

Question 18

A data scientist is building a GenAI application that uses the OpenAI SDK to call a foundation model hosted on Databricks Mosaic AI Model Serving. They want to authenticate the SDK calls securely without hardcoding credentials. Which authentication method should they use?

Accepted Answer

OAuth token from a service principal. OAuth tokens from a service principal are the most secure and recommended way to authenticate SDK calls to Databricks Mosaic AI Model Serving. They are short-lived, can be scoped to specific resources, and are automatically rotated. This method avoids hardcoding credentials and provides better security than PATs or secrets, making it ideal for production GenAI applications.

Answer

Azure Active Directory token

Answer

Personal access token (PAT) stored in environment variables

Answer

Databricks secrets with the API token

Question 19

A team is building a generative AI application that uses Databricks Model Serving to host a fine-tuned LLM. They need to implement safeguards to prevent the model from generating harmful or biased content. Which TWO approaches should they use to mitigate this risk? (Choose two.)

Accepted Answer

Add a pre-processing step that filters user prompts for toxic language before sending them to the model.. Pre-processing prompts to filter toxic inputs and fine-tuning on safe data are both proactive measures that reduce the chance of harmful outputs. Pre-processing enforces input policies, while fine-tuning shapes the model's behavior. Together, they provide a robust defense.

Answer

Increase the temperature parameter to make the model's outputs more diverse and less likely to repeat harmful patterns.

Answer

Implement a post-processing filter that scans model outputs for toxic or biased language and blocks them.

Answer

Use a larger model with more parameters, as larger models are inherently less biased.

Question 20

A GenAI engineer has a retrieval-augmented generation chain built with LangChain and hosted on Databricks. The chain's retriever is a Databricks Vector Search index, and the LLM is a Foundation Model API endpoint. The engineer wants to use MLflow's evaluate function to score the chain on a labeled evaluation dataset and then compare runs across prompt revisions. Which approach should the engineer take to instrument the chain so MLflow can automatically capture retriever outputs and generate evaluation metrics?

Accepted Answer

Wrap the chain with mlflow.langchain.log_model and pass the model URI to mlflow.evaluate as the model argument.. Logging the LangChain chain with the LangChain flavor produces an MLflow model that mlflow.evaluate recognizes, letting the evaluator run the chain end to end and compute retrieval and generation metrics against the labeled dataset. This preserves the retriever and LLM components so prompt revisions can be compared across runs in the Experiments UI without custom metric code.

Answer

Create a Databricks Vector Search endpoint and pass its name to mlflow.evaluate as the model argument.

Answer

Write a custom Python function that calls the chain and pass it to mlflow.evaluate as a pyfunc with a signature.

Answer

Register the chain in Unity Catalog as a model version and pass the fully qualified model name to mlflow.evaluate.