1Z0-184-25 Question 132
Single answerYour team is developing a generative AI chatbot on Oracle Cloud Infrastructure to assist customers with troubleshooting product issues. You have a large corpus of internal technical documents stored in an OCI Object Storage bucket, and you want to use retrieval-augmented generation (RAG) to ensure that the chatbot� responses adapt to newly published product updates. Which approach best demonstrates an effective RAG strategy in this scenario?
- A
Generate vector embeddings from the documents, store them in a specialized index, retrieve relevant content based on user queries, and then feed that content into the LLM to refine its responses.
- B
Upload all documents unprocessed and rely solely on the LLM� base training, assuming its internal parameters can handle new knowledge seamlessly.
- C
Append every document verbatim to the prompt for every user query to make sure the LLM has the full dataset available whenever it generates a response.
- D
Extract only keywords from the documents and feed those keywords as prompts to the LLM, hoping it will generate context from them.
Show answer and explanation
Correct answer: A
Explanation
Retrieval-augmented generation (RAG) helps large language models incorporate up-to-date and domain-specific knowledge by dynamically retrieving segments of relevant text. In OCI, best practices typically involve generating embeddings of the data, storing them in a searchable vector store or index, and feeding only the most pertinent text into the model at inference time. For more information, refer to Oracle� documentation on implementing AI-based solutions with data retrieval and context injection.
- A. Correct.
Correct. This approach captures domain-specific knowledge by embedding the documents, storing them in a vector index, and retrieving only the most relevant passages. That content is then used as context for the LLM� response, ensuring the chatbot incorporates up-to-date technical details from the repository.
- B. Incorrect.
Incorrect. An LLM� internal parameters, once trained, are not automatically aware of newly published content. RAG addresses this gap by dynamically retrieving relevant information from external sources.
- C. Incorrect.
Incorrect. Appending every document in full to each prompt is inefficient and may exceed maximum token limits. It also increases latency and costs without guaranteeing the best context for each query.
- D. Incorrect.
Incorrect. Providing only keywords often omits crucial context. RAG depends on retrieving meaningful text segments to guide the LLM, not merely short descriptors that could lead to incomplete or inaccurate answers.