Enhancing Content Quality In Vertex Ai: The Impact of System Instructions, Prompts, and Parameters

This research explores different ways to configure AI when creating educational content with Vertex AI in our AI2Learn project. We looked at how different settings affect the learning summaries and questions the system generates when trying to personalise student learning. We tested Vertex AI's Gemini 2.0 Flash model with different instructions and settings. We then evaluated the generated content based on how clear, complete, relevant and intellectually challenging it was. We found that being very specific in our instructions resulted in better educational content. We got the best results with an intermediate temperature (1) and reasonable word limits. We also found benefits in trying different instructional styles to get different types of explanations and assessment questions.