The point of a Retrieval-Augmented Generation system is to provide infomation that is of a narrow focus. Now can you just design a specialized LLM that would so the same thing. Yes but it would cost more money and take more time to train it then to use a RAG setup where you can start with a LLM with a basic frame work then add a knowledge base or whatever your cloud provider calls it. The knowledge base is smaller , easier to manage, more focused on a topic that applies to the particular end user. So it could be a knowledge base of DND manuals or a medical records for a single medical practics. Its is easier and then cost less to swap out knowledge bases or add infomation to a RAG system then retrain a LLM.
Lets go over what this website is and what it isn't. I did change it to look less like a shopping site and focus mote on just the operation of the chatbots. The Chatbots are not as snart as a general Large language model (LLM) like Chatgpt but are more focused like a customer service bot. Both of these chatbots were built using the Amazon Titan LLM and a Retrieval-Augmented Generation (RAG)system. Their focus is on the subjects of the role playing systems Shadowrun and Advanced Dungeons and Dragon. The site is hosted on AWS S3 and uses AWS bedrock to provide the knowledge bases that are the core of the Rag system which powers the chatbots. AWS Lex is the conversational interface that wraps around it and allows the user to use text or their own voice to speak to the chatbot. Finally, there is route 53 providing DNS routing and Domain services that give the site the easy to remember name of fxartlabs.com.