From LLM to Conversational Agent: A Memory Enhanced Architecture with Fine-Tuning of Large Language Models