NCA-GENM Question 119
Select 3A company is building a customer support chatbot using a large language model (LLM). To ensure the chatbot provides accurate answers that align with the company’s internal documents, they decide to implement a retrieval-augmented generation (RAG) workflow. Which of the following steps are essential for implementing the RAG workflow in this scenario?
- A
Index the company's internal documents into a vector database for efficient retrieval.
- B
Train the LLM from scratch using the company's internal documents.
- C
Enable the LLM to retrieve the most relevant documents from the vector database during inference.
- D
Fine-tune the LLM using customer interaction data to improve conversational tone.
- E
Combine retrieved documents with the LLM's generated responses to ensure factual accuracy.
Show answer and explanation
Correct answers: A, C, E
Explanation
The RAG workflow combines retrieval of relevant documents with the generative capabilities of an LLM to enhance its outputs with factual, domain-specific information. Indexing documents into a vector database (1), enabling retrieval during inference (3), and combining retrieved data with generated responses (5) are critical steps in implementing RAG. Training or fine-tuning the LLM is not inherently required for RAG, as pre-trained models can be leveraged effectively.
- A. Correct.
This is correct. Indexing the company's internal documents into a vector database is a key step in the RAG workflow to enable efficient retrieval of relevant information.
- B. Incorrect.
This is incorrect. Training an LLM from scratch is not necessary for RAG; pre-trained LLMs are typically used with retrieval mechanisms for domain-specific tasks.
- C. Correct.
This is correct. Enabling the LLM to retrieve relevant documents from the vector database during inference is essential for ensuring responses are grounded in the company's internal knowledge.
- D. Incorrect.
This is incorrect. While fine-tuning can improve conversational tone, it is not specific to the RAG workflow and is optional for this use case.
- E. Correct.
This is correct. Combining retrieved documents with the LLM's responses helps ensure that the generated output is factually accurate and aligned with the company's internal knowledge.