Databricks Databricks-Generative-AI-Engineer-Associate Databricks Certified Generative AI Engineer Associate Exam Practice Test

The task is to set up a Databricks Vector Search index for news articles, supporting queries like “monster truck news around January 5th, 1992,” with minimal effort. The index must filter by topic and a 10-day date range. Let’s evaluate the options.

Option A: Split articles by 10-day blocks and return the block closest to the query

Pre-splitting articles into 10-day blocks requires significant preprocessing and index management (e.g., one index per block). It’s effort-intensive and inflexible for dynamic date ranges.

Databricks Reference:"Static partitioning increases setup complexity; metadata filtering is preferred"("Databricks Vector Search Documentation").

Option B: Include metadata columns for article date and topic to support metadata filtering

Adding date and topic as metadata in the Vector Search index allows dynamic filtering (e.g., date ± 5 days, topic = “monster truck”) at query time. This leverages Databricks’ built-in metadata filtering, minimizing setup effort.

Databricks Reference:"Vector Search supports metadata filtering on columns like date or category for precise retrieval with minimal preprocessing"("Vector Search Guide," 2023).

Option C: Pass the query directly to the vector search index and return the best articles

Passing the full query (e.g., “Tell me about monster truck news around January 5th, 1992”) to Vector Search relies solely on embeddings, ignoring structured filtering for date and topic. This risks inaccurate results without explicit range logic.

Databricks Reference:"Pure vector similarity may not handle temporal or categorical constraints effectively"("Building LLM Applications with Databricks").

Option D: Create separate indexes by topic and add a classifier model to appropriately pick the best index

Separate indexes per topic plus a classifier model adds significant complexity (index creation, model training, maintenance), far exceeding “least effort.” It’s overkill for this use case.

Databricks Reference:"Multiple indexes increase overhead; single-index with metadata is simpler"("Databricks Vector Search Documentation").

Conclusion: Option B is the simplest and most effective solution, using metadata filtering in a single Vector Search index to handle date ranges and topics, aligning with Databricks’ emphasis on efficient, low-effort setups.

The Generative AI Engineer needs to follow a methodical pipeline to build and deploy a Retrieval-Augmented Generation (RAG) application. The steps outlined in optionBaccurately reflect this process:

Ingest documents from a source: This is the first step, where the engineer collects documents (e.g., technical regulations) that will be used for retrieval when the application answers user questions.

Index the documents and save to Vector Search: Once the documents are ingested, they need to be embedded using a technique like embeddings (e.g., with a pre-trained model like BERT) and stored in a vector database (such as Pinecone or FAISS). This enables fast retrieval based on user queries.

User submits queries against an LLM: Users interact with the application by submitting their queries. These queries will be passed to the LLM.

LLM retrieves relevant documents: The LLM works with the vector store to retrieve the most relevant documents based on their vector representations.

LLM generates a response: Using the retrieved documents, the LLM generates a response that is tailored to the user's question.

Evaluate model: After generating responses, the system must be evaluated to ensure the retrieved documents are relevant and the generated response is accurate. Metrics such as accuracy, relevance, and user satisfaction can be used for evaluation.

Deploy it using Model Serving: Once the RAG pipeline is ready and evaluated, it is deployed using a model-serving platform such as Databricks Model Serving. This enables real-time inference and response generation for users.

By following these steps, the Generative AI Engineer ensures that the RAG application is both efficient and effective for the task of answering technical regulation questions.

In this scenario, the chatbot must avoid answering political questions and instead provide a standard message for such inquiries. Implementing aSafety Guardrailis the appropriate solution for this:

What is a Safety Guardrail?Safety guardrails are mechanisms implemented in Generative AI systems to ensure the model behaves within specific bounds. In this case, it ensures the chatbot does not answer politically sensitive or irrelevant questions, which aligns with the business rules.

Preventing Responses to Political Questions:The Safety Guardrail is programmed to detect specific types of inquiries (like political questions) and prevent the model from generating responses outside its intended domain. When such queries are detected, the guardrail intervenes and provides a pre-defined response: “Sorry, I cannot answer that. I am a chatbot that can only answer questions around insurance.”

How It Works in Practice:The LLM system can include aclassification layeror trigger rules based on specific keywords related to politics. When such terms are detected, the Safety Guardrail blocks the normal generation flow and responds with the fixed message.

Why Other Options Are Less Suitable:

B (Security Guardrail): This is more focused on protecting the system from security vulnerabilities or data breaches, not controlling the conversational focus.

C (Contextual Guardrail): While context guardrails can limit responses based on context, safety guardrails are specifically about ensuring the chatbot stays within a safe conversational scope.

D (Compliance Guardrail): Compliance guardrails are often related to legal and regulatory adherence, which is not directly relevant here.

Therefore, aSafety Guardrailis the right framework to ensure the chatbot only answers insurance-related queries and avoids political discussions.

In the context of designing a chatbot to engage users on a gaming platform,diversity of responses(option B) is a key metric to increase user engagement and retention. Here’s why:

Diverse and Engaging Interactions:A chatbot that provides varied and interesting responses will keep users engaged, especially in an interactive environment like a gaming platform. Gamers typically enjoy dynamic and evolving conversations, anddiversity of responseshelps prevent monotony, encouraging users to interact more frequently with the bot.

Increasing Retention:By offering different types of responses to similar queries, the chatbot can create a sense of novelty and excitement, which enhances the user’s experience and makes them more likely to return to the platform.

Why Other Options Are Less Effective:

A (Randomness): Random responses can be confusing or irrelevant, leading to frustration and reducing engagement.

C (Lack of Relevance): If responses are not relevant to the user’s queries, this will degrade the user experience and lead to disengagement.

D (Repetition of Responses): Repetitive responses can quickly bore users, making the chatbot feel uninteresting and reducing the likelihood of continued interaction.

Thus,diversity of responses(option B) is the most effective way to keep users engaged and retain them on the platform.

The task involves building a question-answering chatbot for an automotive company using a database of vehicle-related documents. The chatbot must efficiently process customer inquiries and provide accurate responses. Let’s evaluate each component to determine which isnotuseful, per Databricks Generative AI Engineer principles.

Option A: Response-generating LLM

An LLM is essential for generating natural language responses to customer queries based on retrieved information. This is a core component of any chatbot.

Databricks Reference:"The response-generating LLM processes retrieved context to produce coherent answers"("Building LLM Applications with Databricks," 2023).

Option B: Invite users to submit long, rather than concise, questions

Encouraging long questions is a user interaction design choice, not a technical component of the chatbot’s architecture. Moreover, long, verbose questions can complicate intent detection and retrieval, reducing efficiency and accuracy—counter to best practices for chatbot design. Concise questions are typically preferred for clarity and performance.

Databricks Reference: While not explicitly stated, Databricks’ "Generative AI Cookbook" emphasizes efficient query processing, implying that simpler, focused inputs improve LLM performance. Inviting long questions doesn’t align with this.

Option C: Vector database

A vector database stores embeddings of the vehicle documents, enabling fast retrieval of relevant information via semantic search. This is critical for a question-answering system with a large document corpus.

Databricks Reference:"Vector databases enable scalable retrieval of context from large datasets"("Databricks Generative AI Engineer Guide").

Option D: Embedding model

An embedding model converts text (documents and queries) into vector representations for similarity search. It’s a foundational component for retrieval-augmented generation (RAG) in chatbots.

Databricks Reference:"Embedding models transform text into vectors, facilitating efficient matching of queries to documents"("Building LLM-Powered Applications").

Conclusion: Option B is not a usefulcomponentin building the chatbot. It’s a user-facing suggestion rather than a technical building block, and it could even degrade performance by introducing unnecessary complexity. Options A, C, and D are all integral to a Databricks-aligned chatbot architecture.

The task is to choose between LSH and HNSW for a vector database index, prioritizing semantic accuracy. The evaluation must assess how well each method retrieves semantically relevant results. Let’s evaluate the options.

Option A: Compare the cosine similarities of the embeddings of returned results against those of a representative sample of test inputs

Cosine similarity measures semantic closeness between vectors, directly assessing retrieval accuracy in a vector database. Comparing returned results’ embeddings to test inputs’ embeddings evaluates how well LSH or HNSW preserves semantic relationships, aligning with the priority.

Databricks Reference:"Cosine similarity is a standard metric for evaluating vector search accuracy"("Databricks Vector Search Documentation," 2023).

Option B: Compare the Bilingual Evaluation Understudy (BLEU) scores of returned results for a representative sample of test inputs

BLEU evaluates text generation (e.g., translations), not vector retrieval accuracy. It’s irrelevant for indexing performance.

Databricks Reference:"BLEU applies to generative tasks, not retrieval"("Generative AI Cookbook").

Option C: Compare the Recall-Oriented-Understudy for Gisting Evaluation (ROUGE) scores of returned results for a representative sample of test inputs

ROUGE is for summarization evaluation, not vector search. It doesn’t measure semantic accuracy in retrieval.

Databricks Reference:"ROUGE is unsuited for vector database evaluation"("Building LLM Applications with Databricks").

Option D: Compare the Levenshtein distances of returned results against a representative sample of test inputs

Levenshtein distance measures string edit distance, not semantic similarity in embeddings. It’s inappropriate for vector-based retrieval.

Databricks Reference: No specific support for Levenshtein in vector search contexts.

Conclusion: Option A (cosine similarity) is the correct approach, directly evaluating semantic accuracy in vector retrieval, as recommended by Databricks for Vector Search assessments.

The LLM classifies mushroom species accurately but includes unwanted reasoning text, and the engineer wants only the label. Let’s assess how to control output format effectively.

Option A: Use few shot prompting to instruct the model on expected output format

Few-shot prompting provides examples (e.g., input: description, output: label). It can work but requires crafting multiple examples, which is effort-intensive and less direct than a clear instruction.

Databricks Reference:"Few-shot prompting guides LLMs via examples, effective for format control but requires careful design"("Generative AI Cookbook").

Option B: Use zero shot prompting to instruct the model on expected output format

Zero-shot prompting relies on a single instruction (e.g., “Return only the label”) without examples. It’s simpler than few-shot but may not consistently enforce succinctness if the LLM’s default behavior is verbose.

Databricks Reference:"Zero-shot prompting can specify output but may lack precision without examples"("Building LLM Applications with Databricks").

Option C: Use zero shot chain-of-thought prompting to prevent a verbose output format

Chain-of-Thought (CoT) encourages step-by-step reasoning, which increases verbosity—opposite to the desired outcome. This contradicts the goal of label-only output.

Databricks Reference:"CoT prompting enhances reasoning but often results in detailed responses"("Databricks Generative AI Engineer Guide").

Option D: Use a system prompt to instruct the model to be succinct in its answer

A system prompt (e.g., “Respond with only the species label, no additional text”) sets a global instruction for the LLM’s behavior. It’s direct, reusable, and effective for controlling output style across queries.

Databricks Reference:"System prompts define LLM behavior consistently, ideal for enforcing concise outputs"("Generative AI Cookbook," 2023).

Conclusion: Option D is the most effective and straightforward action, using a system prompt to enforce succinct, label-only responses, aligning with Databricks’ best practices for output control.

Problem Context: The engineer is developing an LLM-powered live sports commentary platform that needs to provide real-time updates and analyses based on the latest game scores. The critical requirement here is the capability to access and integrate real-time data efficiently with the platform for immediate analysis and reporting.

Explanation of Options:

Option A: DatabricksIQ: While DatabricksIQ offers integration and data processing capabilities, it is more aligned with data analytics rather than real-time feature serving, which is crucial for immediate updates necessary in a live sports commentary context.

Option B: Foundation Model APIs: These APIs facilitate interactions with pre-trained models and could be part of the solution, but on their own, they do not provide mechanisms to access real-time game scores.

Option C: Feature Serving: This is the correct answer as feature serving specifically refers to the real-time provision of data (features) to models for prediction. This would be essential for an LLM that generates analyses based on live game data, ensuring that the commentary is current and based on the latest events in the sport.

Option D: AutoML: This tool automates the process of applying machine learning models to real-world problems, but it does not directly provide real-time data access, which is a critical requirement for the platform.

Thus,Option C(Feature Serving) is the most suitable tool for the platform as it directly supports the real-time data needs of an LLM-powered sports commentary system, ensuring that the analyses and updates are based on the latest available information.

When working within a Databricks notebook, several high-level SDKs are "Databricks-aware." TheMLflow Deployments SDK(C) and theDatabricks Python SDK(D) are designed to automatically look for the DATABRICKS_HOST and DATABRICKS_TOKEN environment variables provided by the notebook context. TheOpenAI client(A), when configured for Databricks via Mosaic AI Gateway, also typically handles authentication via workspace integration in recent versions. However, theREST API via the requests library(B) is a generic Python HTTP client. It has no intrinsic knowledge of the Databricks environment. To use it, an engineer must manually extract the token (e.g., via dbutils.notebook.entry_point...) and explicitly pass it in the Authorization: Bearer header of the request. Without this manual step, the requests library will fail with a 401 Unauthorized error.

To build an LLM-powered system that accesses up-to-date news articles and stock prices, the best approach is tocreate an agentthat has access to specific tools (option D).

Agent with SQL and Web Search Capabilities:By using an agent-based architecture, the LLM can interact with external tools. The agent can query Delta tables (for up-to-date stock prices) via SQL and perform web searches to retrieve the latest news articles. This modular approach ensures the system can access both structured (stock prices) and unstructured (news) data sources dynamically.

Why This Approach Works:

SQL Queries for Stock Prices: Delta tables store stock prices, which the agent can query directly for the latest data.

Web Search for News: For news articles, the agent can generate search queries and retrieve the most relevant and recent articles, then pass them to the LLM for processing.

Why Other Options Are Less Suitable:

A (Summarizing News for Stock Prices): This convoluted approach would not ensure accuracy when retrieving stock prices, which are already structured and stored in Delta tables.

B (Stock Price Volatility Queries): While this could retrieve relevant information, it doesn't address how to obtain the most up-to-date news articles.

C (Vector Store): Storing news articles and stock prices in a vector store might not capture the real-time nature of stock data and news updates, as it relies on pre-existing data rather than dynamic querying.

Thus, using an agent with access to both SQL for querying stock prices and web search for retrieving news articles is the best approach for ensuring up-to-date and accurate responses.

In a Retrieval-Augmented Generation (RAG) system, the key to providing relevant responses lies in the quality of the retrieved context. Here’s why option A is the most appropriate solution:

Context Relevance:The RAG model generates answers based on retrieved documents or context. If the retrieved information is about an irrelevant product, it suggests that the retrieval step is failing to select the right context. The Generative AI Engineer must first assess the quality of what is being retrieved and ensure it is pertinent to the query.

Vector Search and Embedding Similarity:RAG typically uses vector search for retrieval, where embeddings of the query are matched against embeddings of product descriptions. Assessing thesemantic similarity searchprocess ensures that the closest matches are actually relevant to the query.

Fine-tuning the Retrieval Process:By improving theretrieval quality, such as tuning the embeddings or adjusting the retrieval strategy, the system can return more accurate and relevant product information.

Why Other Options Are Less Suitable:

B (Caching FAQs): Caching can speed up responses for frequently asked questions but won’t improve the relevance of the retrieved content for less frequent or new queries.

C (Use a Different LLM): Changing the LLM only affects the generation step, not the retrieval process, which is the core issue here.

D (Different Semantic Search Algorithm): This could help, but the first step is to evaluate the current retrieval context before replacing the search algorithm.

Therefore, improving and assessing the quality of the retrieved context (option A) is the first step to fixing the issue of irrelevant product information.

The most effective way to preprocess prompts using custom code is to write a custom model, such as anMLflow PyFunc model. Here’s a breakdown of why this is the correct approach:

MLflow PyFunc Models:MLflow is a widely used platform for managing the machine learning lifecycle, including experimentation, reproducibility, and deployment. APyFuncmodel is a generic Python function model that can implement custom logic, which includes preprocessing prompts.

Preprocessing Prompts:Preprocessing could include various tasks like cleaning up the user input, formatting it according to specific rules, or augmenting it with additional context before passing it to the LLM. Writing this preprocessing as part of a PyFunc model allows the custom code to be managed, tested, and deployed easily.

Modular and Reusable:By separating the preprocessing logic into a PyFunc model, the system becomes modular, making it easier to maintain and update without needing to modify the core LLM or retrain it.

Why Other Options Are Less Suitable:

A (Modify LLM’s Internal Architecture): Directly modifying the LLM's architecture is highly impractical and can disrupt the model’s performance. LLMs are typically treated as black-box models for tasks like prompt processing.

B (Avoid Custom Code): While it’s true that LLMs haven't been explicitly trained with preprocessed prompts, preprocessing can still improve clarity and alignment with desired input formats without confusing the model.

C (Postprocessing Outputs): While postprocessing the output can be useful, it doesn't address the need for clean and well-formatted inputs, which directly affect the quality of the model's responses.

Thus, using an MLflow PyFunc model allows for flexible and controlled preprocessing of prompts in a scalable way, making it the most effective method.

To understand how a typical RAG-enabled customer-facing chatbot processes a user's question, let’s go through the correct sequence as depicted in the diagram and explained in option A:

Embedding Model (1):The first step involves the user's question being processed through an embedding model. This model converts the text into a vector format that numerically represents the text. This step is essential for allowing the subsequent vector search to operate effectively.

Vector Search (2):The vectors generated by the embedding model are then used in a vector search mechanism. This search identifies the most relevant documents or previously answered questions that are stored in a vector format in a database.

Context-Augmented Prompt (3):The information retrieved from the vector search is used to create a context-augmented prompt. This step involves enhancing the basic user query with additional relevant information gathered to ensure the generated response is as accurate and informative as possible.

Response-Generating LLM (4):Finally, the context-augmented prompt is fed into a response-generating large language model (LLM). This LLM uses the prompt to generate a coherent and contextually appropriate answer, which is then delivered as the final output to the user.

Why Other Options Are Less Suitable:

B, C, D: These options suggest incorrect sequences that do not align with how a RAG system typically processes queries. They misplace the role of embedding models, vector search, and response generation in an order that would not facilitate effective information retrieval and response generation.

Thus, the correct sequence isembedding model, vector search, context-augmented prompt, response-generating LLM, which is option A.

Problem Context: The engineer needs a cost-effective deployment strategy for an LLM application with relatively low request volume.

Explanation of Options:

Option A: Switching to external models may not provide the required control or integration necessary for specific application needs.

Option B: Using a pay-per-token model is cost-effective, especially for applications with variable or low request volumes, as it aligns costs directly with usage.

Option C: Changing to a model with fewer parameters could reduce costs, but might also impact the performance and capabilities of the application.

Option D: Manually throttling requests is a less efficient and potentially error-prone strategy for managing costs.

OptionBis ideal, offering flexibility and cost control, aligning expenses directly with the application's usage patterns.

According to Generative AI engineering standards for Retrieval-Augmented Generation (RAG), chunking strategy is a critical optimization variable. Setting the chunk size to match the model's maximum context window (e.g., 4k or 8k tokens) is a poor practice and should be discarded. Large chunks introduce significant "noise" into the LLM's context, as only a small portion of a massive chunk usually contains the answer to a specific query. This leads to the "lost in the middle" phenomenon where LLMs struggle to extract relevant information from bloated contexts. Furthermore, large chunks reduce the precision of the vector search. Standard best practices involve using smaller, semantically meaningful chunks (typically 256–512 tokens) with overlap to maintain context. In contrast, metadata filtering (B) is essential for narrowing searches to specific car years, fine-tuning embeddings (C) improves retrieval accuracy for domain-specific technical terms, and few-shot examples (D) guide the LLM's output format and tone.

The system answers questions on unfolding news topics using articles and social media, with a concern about toxic outputs from toxic inputs. A guardrail must limit toxicity in the LLM’s responses. Let’s evaluate the options.

Option A: Use only approved social media and news accounts to prevent unexpected toxic data from getting to the LLM

Curating input sources (e.g., verified accounts) reduces exposure to toxic content at the data ingestion stage, directly limiting toxic outputs. This is a proactive guardrail aligned with data quality control.

Databricks Reference:"Control input data quality to mitigate unwanted LLM behavior, such as toxicity"("Building LLM Applications with Databricks," 2023).

Option B: Implement rate limiting

Rate limiting controls request frequency, not content quality. It prevents overload but doesn’t address toxicity in social media inputs or outputs.

Databricks Reference: Rate limiting is for performance, not safety:"Use rate limits to manage compute load"("Generative AI Cookbook").

Option C: Reduce the amount of context items the system will include in consideration for its response

Reducing context might limit exposure to some toxic items but risks losing relevant information, and it doesn’t specifically target toxicity. It’s an indirect, imprecise fix.

Databricks Reference: Context reduction is for efficiency, not safety:"Adjust context size based on performance needs"("Databricks Generative AI Engineer Guide").

Option D: Log all LLM system responses and perform a batch toxicity analysis monthly

Logging and analyzing responses is reactive, identifying toxicity after it occurs rather than preventing it. Monthly analysis doesn’t limit real-time toxic outputs.

Databricks Reference: Monitoring is for auditing, not prevention:"Log outputs for post-hoc analysis, but use input filters for safety"("Building LLM-Powered Applications").

Conclusion: Option A is the most effective guardrail, proactively filtering toxic inputs from unverified sources, which aligns with Databricks’ emphasis on data quality as a primary safety mechanism for LLM systems.

The task is to prevent toxic or unsafe responses in an LLM system using the Foundation Model API with minimal effort. Let’s assess the options.

Option A: Host Llama Guard on Foundation Model API and use it to detect unsafe responses

Llama Guard is a safety-focused model designed to detect toxic or unsafe content. Hosting it via the Foundation Model API (a Databricks service) integrates seamlessly with the existing system, requiring minimal setup (just deployment and a check step), and leverages provisioned throughput for performance.

Databricks Reference:"Foundation Model API supports hosting safety models like Llama Guard to filter outputs efficiently"("Foundation Model API Documentation," 2023).

Option B: Add some LLM calls to their chain to detect unsafe content before returning text

Using additional LLM calls (e.g., prompting an LLM to classify toxicity) increases latency, complexity, and effort (crafting prompts, chaining logic), and lacks the specificity of a dedicated safety model.

Databricks Reference:"Ad-hoc LLM checks are less efficient than purpose-built safety solutions"("Building LLM Applications with Databricks").

Option C: Add a regex expression on inputs and outputs to detect unsafe responses

Regex can catch simple patterns (e.g., profanity) but fails for nuanced toxicity (e.g., sarcasm, context-dependent harm), requiring significant manual effort to maintain and update rules.

Databricks Reference:"Regex-based filtering is limited for complex safety needs"("Generative AI Cookbook").

Option D: Ask users to report unsafe responses

User reporting is reactive, not preventive, and places burden on users rather than the system. It doesn’t limit unsafe outputs proactively and requires additional effort for feedback handling.

Databricks Reference:"Proactive guardrails are preferred over user-driven monitoring"("Databricks Generative AI Engineer Guide").

Conclusion: Option A (Llama Guard on Foundation Model API) is the least-effort, most effective approach, leveraging Databricks’ infrastructure for seamless safety integration.

Problem Context: The task is to design responses for a healthcare-focused chatbot that appropriately addresses the urgency of a patient's symptoms.

Explanation of Options:

Option A: Suggesting articles might be suitable for less urgent inquiries but is inappropriate for symptoms that could indicate a serious condition.

Option B: Given the description of severe symptoms like headaches and dizziness, directing the patient to emergency services is prudent. This aligns with medical guidelines that recommend immediate professional attention for such severe symptoms.

Option C: Offering well-wishes does not address the potential seriousness of the symptoms and lacks appropriate action.

Option D: While gathering more information is part of a detailed assessment, the immediate need here suggests a more urgent response.

Given the potential severity of the described symptoms,Option Bis the most appropriate, ensuring the chatbot directs patients to seek urgent care when needed, potentially saving lives.

The problem describes a Retrieval-Augmented Generation (RAG) system for HR policy Q&A where responses are incomplete and unstructured due to the retriever failing to return sufficient context. The engineer has already tried different embedding and response-generating LLMs without success, suggesting the issue lies in the retrieval process—specifically, how documents are chunked and indexed. Let’s evaluate the options.

Option A: Add the section header as a prefix to chunks

Adding section headers provides additional context to each chunk, helping the retriever understand the chunk’s relevance within the document structure (e.g., “Leave Policy: Annual Leave” vs. just “Annual Leave”). This can improve retrieval precision for structured HR policies.

Databricks Reference:"Metadata, such as section headers, can be appended to chunks to enhance retrieval accuracy in RAG systems"("Databricks Generative AI Cookbook," 2023).

Option B: Increase the document chunk size

Larger chunks include more context per retrieval, reducing the chance of missing relevant information split across smaller chunks. For structured HR policies, this can ensure entire sections or rules are retrieved together.

Databricks Reference:"Increasing chunk size can improve context completeness, though it may trade off with retrieval specificity"("Building LLM Applications with Databricks").

Option C: Split the document by sentence

Splitting by sentence creates very small chunks, which could exacerbate the problem by fragmenting context further. This is likely why the current system fails—it retrieves incomplete snippets rather than cohesive policy sections.

Databricks Reference: No specific extract opposes this, but the emphasis on context completeness in RAG suggests smaller chunks worsen incomplete responses.

Option D: Use a larger embedding model

A larger embedding model might improve vector quality, but the question states that experimenting with different embedding models didn’t help. This suggests the issue isn’t embedding quality but rather chunking/retrieval strategy.

Databricks Reference: Embedding models are critical, but not the focus when retrieval context is the bottleneck.

Option E: Fine tune the response generation model

Fine-tuning the LLM could improve response coherence, but if the retriever doesn’t provide complete context, the LLM can’t generate full answers. The root issue is retrieval, not generation.

Databricks Reference: Fine-tuning is recommended for domain-specific generation, not retrieval fixes ("Generative AI Engineer Guide").

Conclusion: Options A and B address the retrieval issue directly by enhancing chunk context—either through metadata (A) or size (B)—aligning with Databricks’ RAG optimization strategies. C would worsen the problem, while D and E don’t target the root cause given prior experimentation.

Problem Context: When using data to train a model, it’s essential to ensure compliance with licensing to avoid legal risks. Legal issues can arise from using data without permission, especially when it comes from third-party sources.

Explanation of Options:

Option A: Reaching out to data curatorsbeforeusing the data is an appropriate action. This allows you to ensure you have permission or understand the licensing terms before starting to use the data in your model.

Option B: Usingoriginal datathat you personally created is always a safe option. Since you have full ownership over the data, there are no legal risks, as you control the licensing.

Option C: Using data that is explicitly labeled with an open license and adhering to the license terms is a correct and recommended approach. This ensures compliance with legal requirements.

Option D: Reaching out to the data curatorsafteryou have already started using the trained model isnot appropriate. If you’ve already used the data without understanding its licensing terms, you may have already violated the terms of use, which could lead to legal complications. It’s essential to clarify the licensing termsbeforeusing the data, not after.

Thus,Option Dis not appropriate because it could expose you to legal risks by using the data without first obtaining the proper licensing permissions.

Problem Context: After receiving positive feedback for the RAG application prototype, the next step is to formally evaluate the system to pinpoint areas for improvement.

Explanation of Options:

Option A: While cosine similarity scores are useful, they primarily measure similarity rather than the overall performance of an RAG system.

Option B: This option provides a systematic approach to evaluation by testing both retrieval and generation components separately. This allows for targeted improvements and a clear understanding of each component's performance, using MLflow’s metrics for a structured and standardized assessment.

Option C: Benchmarking multiple LLMs does not focus on evaluating the existing system’s components but rather on comparing different models.

Option D: Using an LLM as a judge is subjective and less reliable for systematic performance evaluation.

OptionBis the most comprehensive and structured approach, facilitating precise evaluations and improvements on specific components of the RAG system.