MCP: Multi-Tenant Chat Support System with AI Integration

Abstract: Language models have proved their capability of comprehending human utterances. If such models are used for assisting clients in businesses, this poses another challenge for the model user. The challenge arises from the fact that large language models make claims that are attractive and seem correct, but actually are not. This phenomenon is referred to as "hallucination.".

To solve this problem we use something called Retrieval-Augmented Generation. This is how it works: the model looks at information from an area before it answers a question about that information. We made a system called MCP that uses Retrieval-Augmented Generation. MCP is a platform that businesses can use to talk to their customers. The main thing about MCP is that it keeps each businesss information from the other businesses. MCP can also handle a lot of documents, for each business. It even keeps track of how each business is using the MCP system. This way the businesses can see how often they are using MCP to talk to their customers. We tried out MCP. It worked well. The model was much better, at giving answers and did not make things up as much as other models do. This is because it was looking at information before answering questions.

Keywords: Retrieval-Augmented Generation, Multi-Tenant Architecture, Large Language Models, Vector Databases, Customer Support Systems, Semantic Search.