With the growing attention on large language models (LLMs) like GPT-4 and GPT-3.5, the principle "Garbage in, garbage out" emphasizes the importance of quality input in both prompt engineering and fine-tuning. A novel approach to leveraging LLMs involves building a chatbot with a custom knowledge base, using tools like PubNub and Vectara, to answer questions based on proprietary data. The process involves setting up a vector database on Vectara, where data is indexed into vector embeddings for efficient semantic search. The architecture uses PubNub Functions to manage interactions, signaling when the AI is processing and returning results from Vectara's semantic search. This setup allows businesses to utilize secure, internal data for AI-driven insights, with the flexibility to use other vector databases like Pinecone or Weaviate.