What are the key architectural components to create a chatbot based on LLM? Please explain the approach on a high-level?