1Z0-1127-25 Multiple Select — 50 Questions

Question 1

Which THREE components are essential in a typical RAG architecture built on OCI? (Select three.)

Accepted Answer

Vector database (e.g., OCI OpenSearch, Autonomous Database). A typical RAG (Retrieval-Augmented Generation) architecture on OCI requires a vector database such as OCI OpenSearch or Autonomous Database (option A) to store and index the embedded representations of documents and perform similarity search at query time. It also requires an embedding model such as Cohere Embed (option C) to convert both the source documents and the user query into dense vectors that can be compared in that vector store. Finally, it requires a large language model such as Cohere Command (option D) to consume the retrieved context along with the user prompt and generate the grounded natural-language answer. Option B is not essential because a data ingestion pipeline can be implemented with many tools (OCI Data Integration, Object Storage events, LangChain loaders, etc.), and Apache Spark is only one possible choice, not a required component. Option E is not essential either, since a prompt template for system instructions is an implementation detail of prompt engineering rather than a defining architectural component of RAG.

Answer

Data ingestion pipeline with Apache Spark

Answer

Prompt template for system instructions

Question 2

A company is using OCI Generative AI Agents to implement a RAG system for employee onboarding. They want to ensure the agent only answers from the uploaded documents and avoids making up information. Which THREE configuration steps should they take?

Accepted Answer

Configure the agent to use only the knowledge base and disable internet search. Option A is correct because restricting the agent to the knowledge base and disabling internet search prevents the model from pulling in external or unverified information, which is essential for grounding answers strictly in the uploaded onboarding documents. Option B is correct because a preamble (system instruction) that explicitly tells the agent to answer only from the provided context reinforces grounded behavior and reduces hallucination. Option D is correct because a RAG system requires a knowledge base that indexes the onboarding documents so the agent can retrieve relevant, authoritative content to ground its responses. Option C is incorrect because raising the temperature to 2.0 increases randomness and creativity, which would make the agent more likely to generate unsupported or fabricated content. Option E is incorrect because fine-tuning the underlying model is not a required RAG configuration step and does not by itself guarantee answers are limited to the uploaded documents.

Answer

Increase the temperature to 2.0 for more creative responses

Answer

Fine-tune the underlying model on the onboarding documents

Question 3

Which three techniques are commonly used to reduce the risk of prompt injection in LLM applications? (Choose three.)

Accepted Answer

Output filtering.. Output filtering (B) is correct because inspecting and constraining the model's generated response before it reaches the user or downstream system can catch leaked instructions, unsafe tool calls, or exfiltrated data that a prompt injection attempt tried to produce. Input sanitization (D) is correct because stripping or neutralizing suspicious control characters, delimiter sequences, and known injection phrases from user-supplied content reduces the attacker's ability to override or escape the intended prompt structure. Using role-based system prompts (E) is correct because clearly separating trusted system/developer instructions from untrusted user content with distinct roles and delimiters helps the model treat user input as data rather than executable instructions, a core prompt-injection mitigation. Option A is not among the marked correct answers: regex-based prompt validation is brittle and easily bypassed by paraphrasing or encoding, so it is not a commonly recommended primary mitigation. Option C is not correct because increasing temperature raises randomness and creativity, which generally increases the chance of unpredictable or unsafe outputs rather than reducing prompt-injection risk.

Answer

Enabling prompt validation against regex patterns.

Answer

Increasing temperature.

Question 4

In a LangChain RAG pipeline using OCI Generative AI, which THREE components are essential for ingesting documents into a vector store?

Accepted Answer

Text splitter (e.g., RecursiveCharacterTextSplitter). The ingestion phase of a LangChain RAG pipeline requires three essential components. Option B, a text splitter such as RecursiveCharacterTextSplitter, is correct because source documents must be chunked into manageable segments before embedding, since embedding models have token limits and retrieval works best on smaller passages. Option D, a document loader such as PDFLoader, is correct because it is the entry point that reads raw source files and converts them into LangChain Document objects with page_content and metadata. Option E, an embedding model such as OCIGenAIEmbeddings, is correct because it transforms each text chunk into a dense vector that is written into the vector store, which is the defining step of ingestion. Option A, a retriever like vectorstore.as_retriever(), is not part of ingestion; it is used at query time to fetch relevant chunks from an already-populated vector store. Option C, an LLM such as ChatOCIGenAI, is also not required for ingestion because it is used during the generation/answering stage to synthesize a response from retrieved context, not to load, split, or embed documents.

Answer

Retriever (e.g., vectorstore.as_retriever())

Answer

LLM (e.g., ChatOCIGenAI)

Question 5

Which TWO measures can help reduce the risk of generating toxic or unsafe content when using OCI Generative AI Service?

Accepted Answer

Use few-shot prompting with examples that demonstrate safe and appropriate responses.. Option A is correct because few-shot prompting supplies the model with concrete examples of safe, appropriate responses, which steers generation toward the desired behavior and reduces the likelihood of toxic output. Option E is correct because OCI Generative AI Service provides built-in content moderation/filtering that detects and blocks harmful or unsafe content before it is returned to the user. Option B is wrong because disabling monitoring and logging removes visibility and auditability of unsafe outputs rather than reducing risk. Option C is wrong because increasing temperature makes output more random and creative, not more deterministic; lower temperature would increase determinism. Option D is wrong because fine-tuning on unfiltered data can amplify toxic or unsafe patterns instead of mitigating them.

Answer

Disable model monitoring and logging to reduce overhead.

Answer

Increase the temperature parameter to make output more deterministic.

Answer

Fine-tune the model on a large dataset without any safety filtering.

Question 6

An enterprise needs to deploy a custom fine-tuned model for real-time inference with strict latency requirements. They also need to manage costs by paying only for usage. Which three steps are required to achieve this? (Select THREE)

Accepted Answer

Create an endpoint using InferenceClient to point to the dedicated cluster. Option B is correct because creating an endpoint with InferenceClient (or the equivalent endpoint-creation API) is the required step to expose the fine-tuned model on the dedicated cluster so applications can send real-time inference requests to it. Option D is correct because a Dedicated AI Cluster with sufficient model units must be provisioned to provide the dedicated, predictable compute capacity needed to meet strict latency requirements. Option E is correct because the fine-tuned model must actually be hosted/deployed on that Dedicated AI Cluster; provisioning the cluster alone does not serve the model. Option A is not correct because on-demand token-based inference on shared infrastructure does not provide the dedicated capacity guarantees needed for strict latency SLAs. Option C is not correct because deleting the fine-tuning job after deployment is not a required step and does not affect inference costs in the way described; the deployed model and cluster are what incur usage costs.

Answer

Use the on-demand token-based inference on shared infrastructure

Answer

Delete the fine-tuning job after deployment to save costs

Question 7

A developer is troubleshooting an OCI Generative AI inference request that returns a 400 Bad Request error. Which three common causes could result in this error? (Choose three.)

Accepted Answer

Incorrect endpoint URL. Option A is correct because a 400 Bad Request is returned when the request is sent to an incorrect or malformed endpoint URL, such as a wrong region-specific hostname or a mistyped path, so the service cannot route or parse the request. Option C is correct because OCI Generative AI requires specific fields in the JSON body (for example modelId, compartmentId, and the chat or generateText payload); omitting required parameters causes the service to reject the request with 400. Option D is correct because exceeding the model's maximum token limit (input plus output tokens) violates the request constraints and produces a 400 Bad Request rather than a server-side 5xx error. Option B is not correct because an invalid API key typically yields 401 Unauthorized (or 403 Forbidden), not 400 Bad Request. Option E is not correct because network connectivity issues generally manifest as timeouts or connection errors, not an HTTP 400 response from the service.

Answer

Invalid API key in the request header

Answer

Network connectivity issues

Question 8

A prompt engineer is using the ReAct pattern to enable the model to reason and act (e.g., call tools). Which THREE components are essential in the prompt to implement ReAct correctly?

Accepted Answer

A list of available actions or tools with descriptions. Option C is correct because ReAct requires the model to know which external tools or actions it can invoke, so the prompt must enumerate the available actions/tools with descriptions that let the model select the right one. Option D is correct because the scratchpad is the working memory where the model records its intermediate Thought, Action, and Observation steps, enabling the interleaved reasoning-and-acting loop that defines ReAct. Option E is correct because few-shot exemplars demonstrating the Thought/Action/Observation cycle teach the model the exact output format and behavior to follow, which is essential for reliable ReAct execution. Option A is not required: while a stop sequence can help halt generation after an Action so the tool result can be injected, it is an implementation convenience rather than an essential ReAct component. Option B is incorrect because ReAct does not depend on a high temperature; in fact, lower temperatures are typically preferred for deterministic tool selection and structured reasoning.

Answer

A stop sequence after each action

Answer

A system prompt that sets a high temperature

Question 9

Which THREE factors are important when designing a multi-turn conversational agent using OCI Generative AI Agents?

Accepted Answer

Manage the context window size to avoid truncating important earlier messages.. Option B is correct because a multi-turn agent must track the growing token count of the conversation and manage the context window (the model's maximum input tokens) so that earlier turns are summarized, trimmed, or otherwise preserved rather than silently truncated, which would break conversational continuity. Option C is correct because production conversational agents need guardrails — content moderation and intent filtering — to detect and block sensitive topics, prompt-injection, or harmful intents before they reach the model or the user. Option E is correct because session management is what persists conversation history across turns, allowing the agent to reference prior user messages and maintain coherent multi-turn context instead of treating each request as stateless. Option A is not correct because response length should be tuned to the task, not maximized; overly long answers waste tokens, increase latency and cost, and can degrade user experience. Option D is not correct because disabling logging does not meaningfully reduce inference latency or cost and removes the observability needed to debug, audit, and improve the agent.

Answer

Always generate the longest possible response to be thorough.

Answer

Disable logging to reduce latency and cost.

Question 10

A developer is using tree-of-thought prompting to solve a complex planning problem. Which THREE characteristics are typical of tree-of-thought?

Accepted Answer

It explores multiple reasoning branches in parallel. Option A is correct because tree-of-thought (ToT) prompting explicitly expands a problem into multiple candidate reasoning paths at each step, evaluating several 'thoughts' in parallel rather than committing to one. Option C is correct because ToT maintains a search tree over partial solutions, so it can backtrack from dead-end or low-scoring branches and pursue alternatives. Option E is correct because ToT relies on an evaluator or heuristic (e.g., a value estimate or voting score) to rank and prune unpromising branches, keeping the search tractable. Option B is wrong because a single linear chain describes chain-of-thought prompting, not ToT. Option D is wrong because ToT returns the best-scoring path found by search, not necessarily a majority-vote single answer.

Answer

It generates a single linear chain of reasoning

Answer

It always returns a single final answer via majority vote

Question 11

Which THREE are common prompt failures that can degrade model performance?

Accepted Answer

Prompt injection vulnerabilities that allow malicious input to override instructions. Option B is correct because prompt injection is a well-known prompt failure in which untrusted input (e.g., user-supplied text or retrieved documents) contains instructions that override or subvert the system prompt, causing the model to ignore its intended task and degrade performance or safety. Option C is correct because ambiguous instructions leave the model without a single clear interpretation, so it may pick an unintended meaning, produce inconsistent outputs, or fail the task even though it followed the prompt literally. Option D is correct because insufficient context means the model lacks the facts, constraints, or background needed to answer correctly, leading to hallucination, guessing, or off-target responses. Option A is not a prompt failure: output token count is a generation/configuration limit, and short outputs are not inherently degraded—truncation is a parameter issue, not a prompt defect. Option E is not a prompt failure either: temperature is a sampling hyperparameter that controls randomness, so while a high value can reduce determinism, it is a model setting rather than a flaw in the prompt itself.

Answer

Using too few tokens in the output

Answer

Setting temperature too high

Question 12

A company wants to use OCI Generative AI Agents to create a RAG-powered customer support system. Which THREE components are essential for the agent to work?

Accepted Answer

A knowledge base. Option A (a knowledge base) is essential because OCI Generative AI Agents uses the knowledge base to store and retrieve the vectorized chunks of enterprise data that ground the RAG responses. Option C (a data source such as an OCI Object Storage bucket) is required because the agent ingests the source documents from that data source to build and populate the knowledge base. Option D (a base LLM such as Cohere Command R) is necessary because the agent relies on an underlying foundation model to reason over the retrieved context and generate the final natural-language answer. Option B is incorrect because a Dedicated AI Cluster is only needed for hosting custom fine-tuned or dedicated model deployments, not for a standard RAG agent using an on-demand base model. Option E is incorrect because OCI Generative AI Agents automatically handles embedding generation with managed embedding models, so a customer-supplied fine-tuned embedding model is not a required component.

Answer

A Dedicated AI Cluster

Answer

A fine-tuned embedding model

Question 13

An OCI administrator is configuring access control for OCI Generative AI. Which three IAM components are required to allow a group of data scientists to call the GenerateText API? (Choose three.)

Accepted Answer

An IAM group for the data scientists. Option A is correct because IAM authorization in OCI is always granted to groups, not directly to individual users, so the data scientists must be placed in an IAM group that the policy can reference. Option C is correct because a policy statement must grant the verb (for example manage or use) on the ai-services-generative-ai-family resource in the relevant compartment, which is what authorizes calls to the GenerateText API. Option E is correct because OCI policies are scoped to a compartment, so the Generative AI resources must reside in (or be covered by) a compartment for the policy to apply. Option B is incorrect because a local peering gateway is a VCN networking component for connecting VCNs and has nothing to do with IAM API authorization. Option D is incorrect because dynamic groups are used to authorize OCI compute instances and other resources as principals, not human data scientists who authenticate as users.

Answer

A local peering gateway

Answer

A dynamic group

Question 14

A company uses OCI Generative AI to generate product descriptions in XML format. The engineer wants to improve adherence to the XML schema. Which THREE prompt components are most critical? (Select three.)

Accepted Answer

Output format specification: 'Use the following XML schema: ...'. Option C is correct because explicitly specifying the XML schema in the prompt (e.g., 'Use the following XML schema: ...') gives the model the exact structure, element names, and nesting to follow, which is the most direct way to enforce schema adherence. Option D is correct because a clear task instruction such as 'Generate a product description as XML' defines the goal and output type, anchoring the model on producing XML rather than prose. Option E is correct because few-shot examples of valid XML product descriptions demonstrate the schema in practice, letting the model pattern-match correct tags, attributes, and formatting, which strongly improves structured output fidelity. Option A is not appropriate because a temperature of 0.9 increases randomness and creativity, which would reduce deterministic adherence to a strict schema; lower temperature (e.g., 0 or near 0) is preferred for structured output. Option B, while potentially useful for content relevance, is not critical for schema adherence because company background does not constrain XML structure, tags, or validity.

Answer

Setting temperature to 0.9

Answer

Context/background about the company's product line

Question 15

A prompt library manager wants to implement version control for prompt templates used across multiple applications. Which THREE practices should they adopt?

Accepted Answer

Use semantic versioning (e.g., v1.2.3) for prompt templates. Option C is correct because semantic versioning (v1.2.3) gives each prompt template a clear, comparable version identifier so applications can pin or upgrade to specific releases and understand the impact of changes (major/minor/patch). Option D is correct because a changelog records what changed and why, providing an audit trail that helps teams review prompt evolution, roll back problematic changes, and communicate updates across applications. Option E is correct because storing prompt templates in a version control system such as Git provides branching, commit history, diffs, code review, and rollback capabilities, which are the foundation of version control for text artifacts. Option A is not among the correct answers because automated testing, while valuable for quality assurance, is a testing practice rather than a version control practice and does not itself track or manage versions. Option B is not correct because storing prompts only in an application database without history eliminates the ability to track changes, compare revisions, or revert to prior versions, which directly contradicts the goal of implementing version control.

Answer

Automatically test prompts on a fixed set of inputs after each change

Answer

Store prompts only in the application's database without history

Question 16

Which TWO are advantages of using retrieval-augmented generation (RAG) over fine-tuning for incorporating new knowledge?

Accepted Answer

Enables the model to access up-to-date information without retraining. Option B is correct because RAG decouples knowledge from model weights: at inference time the system retrieves relevant documents from an external index and injects them into the prompt, so the model can answer with current information without any retraining or weight updates. Option E is correct because RAG only requires updating the underlying document store/vector index when the corpus changes, avoiding the expensive GPU compute and repeated training runs that fine-tuning would demand for large, frequently changing corpora. Option A is not correct because domain-specific writing style is typically learned through fine-tuning on in-domain text, not through retrieval of facts. Option C is not correct because RAG actually depends on a vector database (or similar retrieval index) to embed and search documents. Option D is not correct because RAG adds a retrieval step and inserts retrieved passages into the context, which generally increases token usage and latency rather than reducing them.

Answer

Better at capturing domain-specific writing style

Answer

Eliminates the need for a vector database

Answer

Reduces token usage and latency compared to fine-tuning

Question 17

Which THREE are essential steps in the prompt engineering process for an LLM?

Accepted Answer

Test the prompt with a variety of input examples. Option A is correct because testing the prompt with a variety of input examples is essential to validate that the prompt generalizes across different cases and to surface edge cases or ambiguities before deployment. Option C is correct because defining the desired output format and constraints (for example, JSON schema, length limits, tone, or required fields) gives the model explicit structure and makes responses consistent and machine-parseable. Option E is correct because prompt engineering is inherently iterative: you evaluate model responses, adjust wording, add examples or constraints, and re-test until the output reliably meets the goal. Options B and D are not part of prompt engineering: fine-tuning on a domain corpus changes model weights through training, and INT8 quantization is a model-compression/deployment optimization, neither of which is an essential step in designing and refining a prompt.

Answer

Fine-tune the model on a domain corpus

Answer

Quantize the model to INT8

Question 18

Which two metrics would you monitor to ensure a generative AI deployment on OCI is operating efficiently? (Choose two.)

Accepted Answer

Request throughput (requests per second). Request throughput (requests per second) (C) is correct because it measures how many inference requests the generative AI endpoint is actually serving per unit time, directly reflecting utilization and capacity of the deployment; low or saturated throughput signals inefficiency or a bottleneck. Average inference latency (D) is correct because it captures the time taken to return a model response, and monitoring it (e.g., via OCI Monitoring metrics on the model deployment endpoint) reveals whether the deployment meets performance expectations or is degraded by resource contention. Together these two operational metrics indicate whether the generative AI service is processing workload efficiently. Number of active users (A) is a usage/adoption metric, not a direct efficiency measure of the deployment. Object Storage bucket size (B) reflects storage consumption, which is unrelated to inference efficiency. Model accuracy on validation set (E) is a model-quality metric measured during training/evaluation, not a runtime efficiency metric.

Answer

Number of active users

Answer

Object Storage bucket size

Answer

Model accuracy on validation set

Question 19

Which THREE factors should be considered when choosing between fine-tuning a model and using a pre-trained model with prompt engineering? (Select three.)

Accepted Answer

Size of available dataset. Option B is correct because fine-tuning typically requires a substantial labeled dataset, whereas prompt engineering can work with few or no examples, so the size of the available dataset directly determines which approach is feasible. Option D is correct because fine-tuning consumes significant compute for training and hosting, while prompt engineering mainly uses inference, making the available budget for compute resources a key decision factor. Option E is correct because if the task demands specialized domain-specific terminology or nuanced jargon, fine-tuning can embed that knowledge into the model, whereas prompt engineering must supply it through carefully crafted prompts. Option A is not a primary factor here because both approaches can be optimized for latency and response time depends more on deployment and model size than on the fine-tune-versus-prompt choice. Option C is not relevant because internet connectivity affects how a model is accessed or deployed, not the fundamental decision between fine-tuning and prompt engineering.

Answer

Required response time

Answer

Internet connectivity

Question 20

Which THREE techniques are commonly used to improve the quality of text generation?

Accepted Answer

Temperature scaling. Temperature scaling (A) is correct because it adjusts the softmax logits before sampling, sharpening the distribution at low temperatures to make high-probability tokens more likely and improving coherence. Top-k sampling (B) is correct because it restricts sampling to the k most probable tokens at each step, cutting off the long tail of unlikely tokens that would otherwise degrade fluency. Beam search (E) is correct because it maintains multiple candidate sequences and selects the highest-scoring one, reducing the risk of locally optimal but globally poor outputs. Greedy decoding (C) is not marked correct because, although it is deterministic and can be adequate, it often produces repetitive or suboptimal text and is not typically framed as a quality-improvement technique. Random sampling (D) is not marked correct because sampling uniformly from the full distribution introduces high variance and frequently yields incoherent output.

Answer

Greedy decoding

Answer

Random sampling