1Z0-1127-25 Building LLM Applications with RAG and Vector Search — 25 Questions

Question 1

A developer is building a RAG chatbot for an internal knowledge base. To ensure the system retrieves the most relevant chunks, what is the best practice for chunking?

Accepted Answer

Use semantic chunking with overlap. Semantic chunking with overlap ensures that each chunk preserves coherent meaning by splitting text at natural boundaries (e.g., sentences or paragraphs) rather than arbitrary token counts. The overlap allows the retriever to capture context that spans chunk boundaries, which is critical for RAG systems to retrieve the most relevant chunks for the LLM to generate accurate answers.

Answer

Use very small chunks

Answer

Use fixed-size chunks without overlap

Answer

Use random-sized chunks

Question 2

A company uses OCI Generative AI to create embeddings for a vector search. They notice high latency in search queries. What is one possible optimization?

Accepted Answer

Use approximate nearest neighbor (ANN) search. To reduce latency in vector search, using approximate nearest neighbor (ANN) search is the standard optimization. ANN algorithms, such as those available in OCI Vector Search (e.g., HNSW, IVF), trade a small amount of accuracy for a significant reduction in query latency compared to exact search. Option A (decrease batch size) affects embedding creation, not search latency. Option C (exact search) typically increases latency due to exhaustive distance computations. Option D (increase embedding dimension) increases computational overhead and latency.

Answer

Decrease batch size for embedding creation

Answer

Use exact search for better accuracy

Answer

Increase the embedding dimension

Question 3

An application uses RAG to answer customer queries, but answers are often incomplete because the retrieved chunks do not contain full context. Which adjustment should the developer make?

Accepted Answer

Increase the number of retrieved chunks. Increasing the number of retrieved chunks (C) directly addresses incomplete answers by providing more context from the vector database. With RAG, the retriever fetches the top-k chunks most similar to the query; raising k increases the breadth of information passed to the LLM, reducing the chance that critical context is missing. This is the most straightforward fix when chunks themselves are correctly sized but the retrieved set lacks full coverage.

Answer

Use a different embedding model

Answer

Increase chunk overlap

Answer

Decrease chunk size

Question 4

A team uses OCI OpenSearch as a vector database for RAG. Some queries return no results despite relevant documents being indexed. What is a likely cause?

Accepted Answer

The number of candidates (k) is set too low. In OCI OpenSearch, the k parameter (number of nearest neighbors) determines how many candidate vectors are retrieved during a k-NN search. If k is set too low, the search may miss relevant documents even if they are indexed, because the algorithm only returns the top-k closest vectors. Increasing k broadens the search space and improves recall.

Answer

The vector index is not refreshed after adding documents

Answer

The query text is too long for the embedding model

Answer

The embedding model is incompatible with the document language

Question 5

A developer uses OCI Generative AI with a custom OCI OpenSearch vector store. The text generation model sometimes hallucinates facts not in the retrieved documents. What is the most effective mitigation?

Accepted Answer

Use prompt engineering to instruct the model to stick to the provided context. Prompt engineering that instructs the model to strictly adhere to the provided context is the most effective mitigation because it directly addresses the root cause of hallucination: the model's tendency to generate information beyond the retrieved documents. By explicitly constraining the model's behavior through the system or user prompt, you reduce the likelihood of fabricated facts without altering the retrieval or generation parameters.

Answer

Use a larger retrieval chunk size

Answer

Increase the temperature

Answer

Decrease the maximum token length

Question 6

Which THREE of the following are likely causes if retrieval returns no results despite documents being indexed in an OCI OpenSearch vector store?

Accepted Answer

The k-NN algorithm is misconfigured (e.g., k=0). Setting k=0 in a k-NN search means requesting zero nearest neighbors, which will always return an empty result set regardless of indexed documents. OCI OpenSearch uses the k-NN plugin where k must be a positive integer (typically ≥1) to retrieve any results; a misconfiguration like k=0 is a common operational error that causes no matches to be returned.

Answer

The embedding model dimension mismatch

Answer

The query embedding is out of distribution

Question 7

Refer to the exhibit. Why did the embedding creation fail?

Accepted Answer

The API call was not properly authenticated. The 401 status code in the exhibit indicates an authentication failure. The API call requires a valid OAuth 2.0 token or API key in the Authorization header; without proper authentication, the embedding creation request is rejected regardless of input length, model availability, or region enablement.

Answer

The input text is too short

Answer

The model ID is not available in the us-ashburn-1 region

Answer

The region is not enabled for the Generative AI service

Question 8

Refer to the exhibit. What is a potential issue with this OCI OpenSearch index template configuration?

Accepted Answer

The space_type in settings (l2) differs from the method's space_type (cosinesimil). The index template's settings specify `space_type: l2` (Euclidean distance), while the method's `space_type` is `cosinesimil` (cosine similarity). This mismatch causes OpenSearch to use inconsistent distance metrics during indexing and search, leading to incorrect similarity calculations and degraded retrieval quality.

Answer

The ef_construction parameter is set too low

Answer

Number of replicas is 0, which provides no redundancy

Answer

The dimension 1024 is too large for the knn_vector type

Question 9

Refer to the exhibit. What is the best action to resolve this error?

Accepted Answer

Reduce the number of retrieved documents. The error is caused by exceeding the context window limit of the LLM when too many retrieved documents are passed to the model. Reducing the number of retrieved documents ensures the total token count stays within the model's maximum context length, preventing truncation or failure. This directly addresses the root cause without sacrificing generation quality or altering chunking strategy.

Answer

Decrease the temperature of the generation model

Answer

Increase the max_tokens parameter for generation

Answer

Use a smaller chunk size for documents

Question 10

A company is building a RAG application for customer support. The knowledge base includes documents in English, Spanish, and French. Which embedding model should they use from OCI Generative AI to ensure accurate retrieval across all languages?

Accepted Answer

cohere.embed-multilingual-light-v3.0. The cohere.embed-multilingual-light-v3.0 model is specifically designed to handle multiple languages, including English, Spanish, and French, ensuring accurate semantic retrieval across all three languages in the RAG application. OCI Generative AI offers this multilingual embedding model to support diverse knowledge bases, whereas single-language models would fail to represent non-English text effectively.

Answer

cohere.generate-english-v2:0

Answer

cohere.embed-english-light-v3.0

Answer

cohere.command-r-plus-v1:0

Question 11

An organization is experiencing low recall in their RAG system. They are using OCI OpenSearch as the vector store with cosine similarity. After reviewing the retrieved chunks, they notice that relevant documents are not being returned. Which configuration change is most likely to improve recall?

Accepted Answer

Reduce the chunk size to capture more granular information.. Low recall means the system is missing relevant documents. Reducing chunk size increases the number of chunks, which can capture more granular information and improve the chance that a relevant passage is retrieved. In OCI OpenSearch with cosine similarity, smaller chunks often lead to higher recall because the vector representation can more precisely match query semantics.

Answer

Use a deterministic ID generator for consistent chunk IDs.

Answer

Increase the chunk size to provide more context per chunk.

Answer

Switch similarity metric from cosine to Euclidean distance.

Question 12

A healthcare company is deploying a RAG application using OCI Generative AI and wants to ensure patient data privacy. They cannot send sensitive data to a public embedding endpoint. Which approach should they take to embed documents while maintaining data residency and security?

Accepted Answer

Provision a dedicated AI cluster in their OCI tenancy to host the embedding model.. Provisioning a dedicated AI cluster in your OCI tenancy ensures that the embedding model runs entirely within your own cloud boundary, so patient data never leaves your tenancy. This approach satisfies strict data residency and security requirements by keeping sensitive data on-premises in the cloud, avoiding any public endpoint exposure. It also allows you to control encryption, network isolation, and compliance with regulations like HIPAA.

Answer

Use the standard OCI Generative AI public endpoint with data encryption in transit.

Answer

Use an external embedding service that complies with HIPAA in a different cloud region.

Answer

Hash the document text before sending to the public embedding endpoint.

Question 13

During a RAG implementation, the response quality degrades because the LLM receives too many irrelevant document chunks. Which technique can best filter out irrelevant chunks before sending them to the LLM?

Accepted Answer

Implement a reranking step using a cross-encoder model.. A reranking step using a cross-encoder model (Option C) directly addresses the problem of irrelevant chunks by scoring each retrieved chunk against the user query with a deep, pairwise similarity computation. Unlike a bi-encoder used in initial retrieval, a cross-encoder evaluates query-chunk pairs jointly, allowing it to reorder or filter out chunks that are semantically irrelevant, thus improving the quality of context fed to the LLM.

Answer

Use a larger LLM for generation, hoping it ignores irrelevant chunks.

Answer

Reduce the top-k retrieval count.

Answer

Increase the chunk size to provide more context.

Question 14

A company is using Oracle Database 23ai AI Vector Search for their RAG pipeline. They notice that similarity search often returns chunks that are semantically unrelated but syntactically similar due to token overlap. Which vector index type should they consider to improve semantic relevance?

Accepted Answer

HNSW index. HNSW (Hierarchical Navigable Small World) indexes are designed to prioritize semantic similarity over exact token overlap by constructing a multi-layer graph that navigates based on vector distance in the embedding space. Unlike IVF-based indexes, which rely on coarse quantization and can be misled by high-frequency token patterns, HNSW's graph-based search inherently captures the global structure of the embedding space, making it more robust to syntactically similar but semantically unrelated chunks.

Answer

IVF_SQ8 index

Answer

IVF_FLAT index

Answer

Use the default index type, which is IVF_FLAT

Question 15

A team is designing a RAG system for legal document review. They want to ensure that the retrieved chunks are contextually coherent and not truncated mid-sentence. Which chunking strategy should they use?

Accepted Answer

Recursive chunking based on sentence boundaries.. Recursive chunking based on sentence boundaries ensures that each chunk ends at a natural sentence boundary, preserving contextual coherence and avoiding mid-sentence truncation. This is critical for legal document review where incomplete sentences can lead to misinterpretation of legal clauses. The recursive approach allows for variable-length chunks that respect linguistic structure, unlike fixed-size or token-level methods.

Answer

Token-level chunking.

Answer

Semantic chunking using document section headers.

Answer

Fixed-size character chunking with overlap.

Question 16

Which TWO of the following are best practices when indexing documents for a RAG application using OCI OpenSearch?

Accepted Answer

Enable chunk overlap to avoid splitting important information across chunks.. Chunk overlap ensures that semantically related content that spans chunk boundaries is preserved, preventing the loss of important context when generating embeddings. In OCI OpenSearch, overlapping chunks help maintain continuity for downstream retrieval, as the retriever can still access the full context even if a sentence or concept is split across two chunks.

Answer

Use the same chunk size for all documents regardless of content type.

Answer

Apply stemming to reduce vocabulary size.

Answer

Remove all stop words from the text before embedding.

Question 17

Which THREE factors should be considered when choosing a vector store for a RAG application in OCI?

Accepted Answer

Capability of hybrid search (vector + keyword).. Hybrid search combines vector similarity search with keyword-based (BM25) retrieval, which significantly improves relevance for RAG applications by handling exact term matches and semantic similarity together. OCI's OpenSearch and AI Vector Search support hybrid search, enabling more accurate context retrieval for LLM prompts.

Answer

The maximum vector dimension supported.

Answer

The CPU architecture of the vector store nodes.

Question 18

A developer sends the above request to the OCI Generative AI API. The response returns an error: 'InvalidParameter: The parameter 'topP' is not supported for this model.' What is the most likely reason?

Accepted Answer

The 'cohere.command-r-plus-v1:0' model does not accept the topP parameter.. The error 'InvalidParameter: The parameter 'topP' is not supported for this model' indicates that the model specified in the request, 'cohere.command-r-plus-v1:0', does not accept the 'topP' parameter. This is because the Cohere Command-R+ model uses a different sampling strategy (e.g., temperature and top-k) and does not expose a 'topP' parameter in the OCI Generative AI API. The API validates parameters against the model's capabilities, and unsupported parameters trigger this specific error.

Answer

The model ID is deprecated.

Answer

The topP parameter value is out of range.

Answer

The JSON request format is incorrect.

Question 19

A developer implements a RAG chatbot using OCI Generative AI with streaming enabled. The chatbot fails to remember earlier conversation turns during a session. What is the most likely cause?

Accepted Answer

The application does not include previous messages in the request.. A RAG chatbot with streaming enabled still requires the application to manage conversation state by including previous messages in each request. The OCI Generative AI streaming endpoint processes each request independently; without explicitly passing the conversation history, the model has no context of prior turns, causing it to fail to remember earlier interactions.

Answer

The max_tokens parameter is set too low.

Answer

The streaming endpoint does not support conversation history.

Answer

The temperature parameter is too high.

Question 20

A data scientist is designing a RAG system with a large vector database (hundreds of millions of documents) and requires high recall accuracy. Which vector search index type should be used in OCI Search with OpenSearch?

Accepted Answer

HNSW (Hierarchical Navigable Small World). HNSW is the correct choice because it provides the best trade-off between high recall and low latency for large-scale vector search. It builds a multi-layer graph structure that enables efficient approximate nearest neighbor search, achieving recall accuracy close to brute-force while scaling to hundreds of millions of documents.

Answer

LSH (Locality Sensitive Hashing)

Answer

Flat (brute-force)

Answer

IVF (Inverted File Index)