1Z0-1127-25 Building LLM Applications with RAG and Vector Search Practice Test 9 — 15 Questions

Question 1

A data scientist is designing a RAG system using OCI Data Science and OCI Generative AI. Which two considerations are critical for optimal retrieval quality? (Choose 2.)

Accepted Answer

Use the same embedding model for both indexing and querying.. Option A is correct because retrieval quality in a RAG pipeline depends on the query embedding and the document embeddings living in the same vector space; using the same embedding model (e.g., the same OCI Generative AI embedding model and version) for both indexing and querying ensures cosine similarity scores are meaningful and comparable. Option D is correct because applying metadata filtering (for example, on attributes such as document type, date, or department stored alongside vectors in a vector store) narrows the candidate set before the vector similarity search, which reduces irrelevant matches and improves precision, especially in large corpora. Option B is not correct because maximizing chunk size can dilute the embedding with unrelated content and exceed the model's context window, hurting retrieval precision rather than helping it. Option C is not correct because fine-tuning the generation model affects answer synthesis, not retrieval quality, and is not a critical retrieval consideration. Option E is not correct because a hierarchical index structure primarily addresses search speed and scalability, not the relevance or quality of retrieved results.

Answer

Increase the chunk size to the maximum allowed to capture more context.

Answer

Fine-tune the generation model on domain-specific data.

Answer

Use a hierarchical index structure for faster search.

Question 2

A small business is building an internal Q&A bot using OCI Generative AI with RAG. They have indexed their product manuals into OCI OpenSearch using a precomputed embedding model. When they test queries, the bot often returns answers that are only partially relevant, and sometimes it cannot find answers for questions that are clearly present in the manuals. The developers suspect the chunking strategy is suboptimal. Currently, they use a fixed chunk size of 512 tokens with no overlap. What should they do to improve retrieval relevance?

Accepted Answer

Add a 20% token overlap between consecutive chunks.. Adding a 20% token overlap between chunks ensures that semantically related content spanning chunk boundaries is not lost, which directly addresses the problem of missing answers for questions present in the manuals. Overlap preserves context continuity, improving retrieval recall and relevance without sacrificing the precision that a fixed-size chunk provides.

Answer

Increase the chunk size to 1024 tokens to include more context.

Answer

Reduce the chunk size to 256 tokens to increase precision.

Answer

Switch to a sentence-based chunking strategy with no overlap.

Question 3

A developer is using OCI Data Science to create a RAG pipeline. They have ingested documents into a vector store using OCI Generative AI's text-embedding model. During testing, they notice that queries return very few results (often 0 or 1) even when the knowledge base contains relevant documents. They have set the top-k parameter to 10. What is the most likely cause?

Accepted Answer

The similarity threshold is set too high, filtering out most results.. In a RAG pipeline using vector search, the similarity threshold parameter filters out any retrieved chunks whose similarity score falls below the configured value. If the threshold is set too high (e.g., 0.9), even semantically relevant documents may score below it, resulting in zero or one result despite top-k being set to 10. This is the most likely cause when top-k is correctly configured but results are unexpectedly sparse.

Answer

The documents were chunked with too small a chunk size, losing key information.

Answer

The embedding model's dimensionality is too low to capture semantic differences.

Answer

The vector search index is not configured with the correct distance metric.

Question 4

A company uses OCI Generative AI's chat endpoint with RAG for customer support. They have observed that the model sometimes generates answers that contradict the retrieved context. The retrieved chunks are correct and relevant, but the model ignores them. What configuration change should they implement first?

Accepted Answer

Add explicit instructions in the system prompt to base answers solely on provided context.. The correct option is C: adding explicit instructions in the system prompt to base answers solely on the provided context. Since the retrieved chunks are already correct and relevant, the problem is that the model is not grounding its output in them, and a system prompt that mandates context-only answering directly addresses this behavior without retraining or changing the retrieval pipeline. Option A (fine-tuning) is unnecessary and costly because the retrieval data is already accurate, and fine-tuning does not guarantee adherence to retrieved context at inference time. Option B (increasing retrieved chunks from 3 to 5) would not help because the issue is not missing or insufficient context, and more chunks could even dilute attention. Option D (changing the embedding model) is irrelevant because retrieval quality is already confirmed to be correct and relevant.

Answer

Fine-tune the generation model on customer support dialogues.

Answer

Increase the number of retrieved chunks from 3 to 5.

Answer

Use a different embedding model for retrieval.

Question 5

An enterprise is deploying a RAG application for compliance document analysis using OCI. They use OCI OpenSearch as the vector store and have millions of documents. Retrieval latency is critical. Currently, a single query takes over 2 seconds. The index uses a flat (brute-force) distance computation. They have considered using approximate nearest neighbor (ANN) algorithms but are unsure about the impact on recall. They need to reduce latency to under 500ms while maintaining high recall. What should they do?

Accepted Answer

Switch to an HNSW algorithm with an appropriate M and ef_search parameters.. Switching to HNSW with appropriate parameters provides fast approximate search with configurable recall. Option A (reducing dimensions) can degrade embedding quality. Option B (reducing shards) may not achieve the required latency reduction. Option D (increasing top-k) would increase latency.

Answer

Use a smaller embedding dimension by truncating the existing embeddings.

Answer

Reduce the number of shards in the OpenSearch index to improve parallelism.

Answer

Increase the top-k parameter to retrieve more candidates then filter.

Question 6

A data scientist is using OCI Data Science to build a RAG system for medical literature. They have a large corpus of PDFs. They used the default OCI Generative AI embedding model and chunked each PDF into 512-character segments with 10% overlap. However, queries about specific drug doses often return incorrect information, even though the correct dose is present in the corpus. Upon inspection, they find that the retrieved chunks often contain partial dose information or miss the context units (e.g., mg vs. mcg). What improvement should they prioritize?

Accepted Answer

Use a semantic chunking strategy that respects document structure (e.g., paragraphs, sections).. The core issue is that fixed-size character chunking (512 characters) breaks medical documents mid-sentence or mid-section, causing critical context like units (mg vs. mcg) to be separated from the numeric dose. Semantic chunking that respects document structure (paragraphs, sections) ensures that dose information and its units remain in the same chunk, preserving the semantic integrity needed for accurate retrieval. This directly addresses the observed problem of partial dose information and missing context units.

Answer

Implement a secondary verification step using a rule-based pattern matcher.

Answer

Increase the chunk overlap to 50% to ensure more context.

Answer

Fine-tune the embedding model on medical text.

Question 7

A company is using OCI Generative AI for a RAG-based code assistant. They index source code repositories into a vector store. Developers report that the assistant often suggests deprecated APIs or outdated code snippets, even though the latest code is in the repository. The index was built a week ago and has not been updated. They plan to set up incremental updates. However, they notice that even after re-indexing the latest commits, the issue persists. What is the most likely oversight?

Accepted Answer

The vector store is not configured to overwrite existing vectors for updated documents.. The most likely oversight is that the vector store is not configured to overwrite existing vectors for updated documents. When re-indexing, if the system appends new vectors without deleting or updating the old ones, the retrieval may still surface deprecated code snippets from the outdated vectors. This leads to the assistant suggesting obsolete APIs despite the latest code being indexed.

Answer

The retrieval top-k is set too low, missing some relevant snippets.

Answer

The chunking strategy splits code at function boundaries, losing import statements.

Answer

The embedding model is not fine-tuned on code; it was trained on natural language.

Question 8

A solutions architect is building a RAG pipeline on OCI Generative AI. The team stores document chunks in an OCI Search with OpenSearch index and wants the vector dimension of the index to match the embeddings produced by the cohere.embed-english-v3.0 model. Which action should the architect take to ensure the index mapping is compatible with the embedding output?

Accepted Answer

Create the index with a knn_vector field whose dimension matches the embedding model's output dimension (for example, 1024 for cohere.embed-english-v3.0).. The k-NN vector field dimension must equal the embedding length produced by the model. Cohere Embed English v3.0 returns 1024-dimensional vectors by default, so declaring a knn_vector with dimension 1024 lets documents index and lets queries run approximate nearest neighbor search. Shard counts and dynamic mapping cannot substitute for matching the declared dimension.

Answer

Set the index's number_of_shards to match the embedding dimension so each shard stores one vector component.

Answer

Enable dynamic mapping and let OpenSearch infer the vector dimension from the first document indexed.

Answer

Store embeddings as base64-encoded strings in a text field and convert them at query time.

Question 9

An AI engineer runs a RAG chatbot on OCI Generative AI. Users report that answers drift off-topic when a single query retrieves many loosely related chunks. The engineer wants to reduce context noise while keeping the most semantically relevant passages. Which approach best addresses this?

Accepted Answer

Apply a similarity score threshold and rerank the retrieved chunks, then pass only the top few to the model.. Off-topic drift from loosely related chunks is a context-quality problem. Applying a similarity threshold discards weak matches and a reranker reorders survivors by relevance to the query, so only the strongest passages reach the model. That reduces noise and keeps the answer grounded, whereas enlarging top_k, changing embedding dimension, or lowering temperature does not improve which chunks are selected.

Answer

Increase the top_k of the vector search so more candidate chunks are passed to the model for grounding.

Answer

Lower the model's temperature to zero so the LLM produces more deterministic answers from the same context.

Answer

Switch the embedding model to one with a larger output dimension so each chunk carries more semantic detail.

Question 10

You are building a RAG assistant on OCI Generative AI. After ingestion, you notice that semantically similar questions phrased differently (for example, 'How do I reset my password?' vs 'What is the procedure to change my login credentials?') return different top-k chunks. You want to improve semantic recall without changing the LLM. Which action should you take?

Accepted Answer

Replace the embedding model with a higher-quality OCI Generative AI embedding model and re-index the corpus.. Semantic recall for paraphrased queries is primarily a function of embedding quality. A stronger OCI embedding model maps semantically equivalent text to nearby vectors, so different phrasings retrieve the same relevant chunks. Chunk overlap, top-k, and temperature do not change the underlying vector representation, so they cannot fix a semantic mismatch between query and document embeddings.

Answer

Lower the temperature of the LLM to 0 so that paraphrased questions produce consistent answers.

Answer

Increase the top-k retrieval parameter from 5 to 50 and pass all chunks to the LLM.

Answer

Increase the chunk overlap from 10% to 20% while keeping the same embedding model and vector index.

Question 11

A RAG application on OCI Generative AI ingests 40,000 technical manuals. Each manual is chunked, embedded with the Cohere Embed model, and stored in an OCI OpenSearch vector index. Users now complain that answers are slow and often cite chunks from unrelated documents. Logs show every query performs an exhaustive scan across all 40,000 vectors. What should you implement to improve both latency and relevance?

Accepted Answer

Create an OpenSearch index with an HNSW-based knn_vector field and query it with approximate nearest neighbor search using the query embedding.. The exhaustive scan across all vectors is the core bottleneck, and the unrelated citations indicate ranking is not using vector similarity effectively. Building an HNSW-backed knn_vector index in OCI OpenSearch and querying with approximate nearest neighbor search reduces comparisons and ranks semantically similar chunks first. The other choices either keep the full scan, weaken semantic precision, or only mask repeated traffic.

Answer

Increase the chunk size to 8,000 tokens so each vector represents a full manual section and fewer vectors are compared per query.

Answer

Enable OCI Generative AI response caching and set a longer time-to-live so repeated queries return faster.

Answer

Switch the embedding model to a larger Cohere model and re-embed all documents nightly to raise retrieval accuracy.

Question 12

A RAG application on OCI Generative AI uses a vector store with 1 million chunks. Users report that answers are correct but slow. Profiling shows the vector search step dominates latency. You must reduce retrieval latency while preserving recall. Which approach should you take?

Accepted Answer

Switch from exact k-nearest-neighbor search to an approximate nearest-neighbor index such as HNSW or IVF.. For large vector stores, exact nearest-neighbor search scans all vectors and becomes the bottleneck. Approximate nearest-neighbor indexes such as HNSW or IVF reduce the number of distance computations by navigating a graph or partitioning the space, cutting latency dramatically while keeping recall high when parameters are tuned. Other options either harm recall, help only a subset of queries, or fail to change the algorithmic cost.

Answer

Reduce the embedding dimensionality by truncating vectors to the first 128 dimensions.

Answer

Move the vector store to a larger compute shape with more memory and CPU.

Answer

Cache the LLM responses for repeated queries and return the cached answer without retrieval.

Question 13

A support team is building a RAG assistant on OCI Generative AI. They want the assistant to answer questions using only the company's internal policy documents stored in an OCI Object Storage bucket. They need to convert those documents into vectors and store them in a vector database for similarity search. Which OCI service should they use to generate the vector representations of the document chunks?

Accepted Answer

OCI Generative AI embedding models, such as Cohere Embed, invoked through the Generative AI inference endpoint.. Generating vector representations from text requires an embedding model, and OCI Generative AI exposes embedding models such as Cohere Embed through its inference endpoint. The other services handle transcription, image analysis, or metadata cataloging, none of which produce text embeddings. Keeping embeddings inside OCI Generative AI also aligns IAM and networking with the generation step of the RAG pipeline.

Answer

OCI Speech to transcribe each policy document and emit vector coordinates for indexing.

Answer

OCI Vision to extract features from document images and store them as embeddings.

Answer

OCI Data Catalog to profile the documents and generate semantic hashes for similarity comparison.

Question 14

A RAG application on OCI Generative AI uses Cohere Embed v3 to embed 50,000 support articles. The team notices that semantically similar articles about 'refund policy' and 'return policy' are not retrieved together, and cosine similarity scores cluster narrowly between 0.78 and 0.82. They want to improve retrieval discrimination without changing the embedding model. Which approach is most appropriate?

Accepted Answer

Apply a hybrid retrieval strategy that combines vector similarity with keyword or BM25 matching. Hybrid retrieval combines dense vector search with sparse lexical signals such as BM25 or keyword matching, so lexically overlapping but semantically similar documents are more likely to be retrieved. This compensates for compressed cosine similarity bands without altering the embedding model, and it is a common pattern in production RAG systems to improve recall for domain terminology and near-synonym phrasing.

Answer

Increase the LLM temperature so the model explores more diverse interpretations of the query

Answer

Switch the distance metric to Euclidean and re-index the embeddings

Answer

Reduce the embedding dimensions to 256 and rebuild the index to increase score variance

Question 15

You are designing a RAG pipeline on OCI Generative AI where documents are updated frequently. Users must see the latest content immediately after an update. The current batch re-indexing runs nightly, causing stale answers during the day. Which design change best meets the freshness requirement?

Accepted Answer

Implement incremental indexing that embeds and upserts changed documents in near real time.. Real-time freshness requires an event-driven, incremental indexing pipeline that embeds and upserts documents as they change. Batch re-indexing, chunk resizing, and scheduled cache invalidation all introduce delays or fail to update the vector store promptly. Incremental indexing minimizes the window between an update and its availability in retrieval, directly satisfying the requirement.

Answer

Reduce the chunk size so that updates affect fewer chunks and propagate faster.

Answer

Add a cache layer that stores LLM responses and invalidates it on a schedule.

Answer

Increase the batch re-indexing frequency to every hour.