Implement RAG using the OCI Generative AI Service
Free Oracle Cloud Infrastructure 2025 Generative AI Professional practice — 6 questions on Implement RAG using the OCI Generative AI Service, with explanations. No sign-up. Full 12-question mixed test →
A company builds a RAG chatbot from a 500-page technical manual. Chunks that are too large exceed the embedding model's token limit, and chunks that are too small lose surrounding context needed for accurate retrieval. Which chunking strategy BEST balances these constraints?
Fixed-size token chunking with overlap keeps chunks within embedding model limits while preserving semantic continuity at chunk boundaries, which is the standard production RAG chunking pattern.
Your RAG corpus embeddings total 200GB, exceeding the memory available on the database server. You need an approximate similarity search index that persists to disk while still providing good query performance. Which Oracle Database 23ai AI Vector Search index type is appropriate?
IVF (Inverted File Flat) indexes in Oracle 23ai AI Vector Search partition vectors into neighbor lists stored on disk, making them suited for datasets too large to fit entirely in memory while still supporting fast approximate search.
You generated document embeddings using cohere.embed-english-light-v3.0 (384 dimensions), but the target VECTOR column in Oracle Database 23ai was declared as VECTOR(1024, FLOAT32). What happens when you attempt to insert these embeddings?
Oracle 23ai enforces the declared VECTOR column dimension strictly; inserting a vector of a different dimension than declared raises an error at insert time rather than silently succeeding.
In a RAG pipeline, why is semantic search using vector embeddings generally preferred over traditional keyword-based search for retrieving relevant context from private enterprise documents?
Semantic search encodes meaning into dense vectors so it can match paraphrased or conceptually related queries to relevant documents even when no exact keyword overlap exists, improving retrieval recall for RAG.
You are building a LangChain-style RAG pipeline on the OCI Generative AI service. The pipeline retrieves the top-k relevant chunks from a vector store, then must combine them with the user's original question before calling the generation model. Which orchestration component performs this combination step?
The prompt template or combination step (often called a 'stuff' or context-injection chain) is responsible for merging retrieved context chunks with the user's query into a single formatted prompt sent to the LLM for generation.
A RAG chatbot built on the OCI Generative AI Service occasionally answers questions using facts not present in the retrieved documents, even when relevant context was successfully retrieved. Which technique BEST reduces this specific failure mode?
Explicit grounding instructions that restrict the model to the supplied context, and direct it to admit when information is missing, directly address hallucination that occurs during the generation step even after correct retrieval.
Ready for the real thing?
The full course has two full-length practice tests, video lessons for every exam domain, hands-on labs and detailed answer explanations.
Start my full course on Udemy →