LLM-as-a-Grader: Practical Insights from Large Language Model for Short-Answer and Report Evaluation