An AI development company is working on an AI-assisted chatbot for a customer, which happens to be an online retail company. The goal is to create an…

The scenario requires a chatbot that can accurately answer questions based on a specific, proprietary knowledge base (company policies) and maintain conversational context. An LLM enhanced with Retrieval-Augmented Generation (RAG) is the best fit. RAG combines a retriever, which fetches relevant information from an external data source (like a vector database of company policies), with a generator (the LLM). This grounds the LLM's responses in factual, up-to-date company data, preventing hallucinations and ensuring policy-related queries are answered correctly. The LLM component handles the conversational flow and session history.

A. A keyword search-based AI is too rudimentary. It lacks the natural language understanding to handle complex or nuanced queries about policies that don't contain exact keywords.

C. An LLM without external data integration would rely solely on its pre-trained knowledge, which is not specific to the retail company's policies and could provide incorrect or fabricated information (hallucinations).

D. This option specifies the source of a model (e.g., Cohere, OpenAI), not the architectural pattern. While such a model would be used, it is the RAG architecture that makes it suitable for this specific task.

1. Oracle Cloud Infrastructure (OCI) Documentation. "Retrieval-Augmented Generation (RAG) in OCI Generative AI". This document states, "RAG enhances the accuracy and relevance of Large Language Model (LLM) generated responses by retrieving information from a specified data source and using it to create a response... RAG is useful for question and answering on your private data." This directly supports using RAG for querying company-specific information.

Source: Oracle Cloud Infrastructure Documentation, Generative AI Service, "Overview of Retrieval-Augmented Generation (RAG)".

2. Oracle Cloud Infrastructure (OCI) Blog. "Build a Q&A solution over your enterprise data using OCI Generative AI". This article details an architecture for a question-answering system using enterprise data. It highlights the problem: "LLMs are trained on public data and don't have access to your private enterprise data." It then presents a RAG-based solution as the way to "ground the LLM with your own data to get more accurate and relevant completions."

Source: Oracle Cloud Infrastructure Blog, AI & Data Science section, "Build a Q&A solution over your enterprise data using OCI Generative AI," published October 24, 2023.

3. Lewis, P., et al. (2020). "Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks." Advances in Neural Information Processing Systems 33. This foundational academic paper introduced the RAG model, demonstrating that it significantly improves performance on knowledge-intensive tasks by combining pre-trained language models with a non-parametric memory (a retriever over a knowledge corpus like Wikipedia). This principle is directly applicable to using a corpus of company policies.

DOI: Available via arXiv:2005.11401. See Section 1 (Introduction) and Section 3 (Methods).