Improving LLM-Powered EDA Assistants with RAFT