Constructing a RAG system tailored for a particular domain or enterprise at a production level poses numerous challenges. These include content curation and preprocessing, text embedding and vector database choice, and assessing the RAG system's performance. In the first part of the talk, we'll delve into the practical considerations while building the AI Assistant, which will be gleaned from data experiments and real-world implementations. The second part will provide more details about the cybersecurity of the AI Assistant itself by addressing how to harden the LLM applications from prompt injection, data poisoning, and other attacks.