Proof-of-concept Google Workspace Add-on using Gemini to analyze and provide feedback on presentations.

The add-on uses the Slides API to extract thumbnails and speaker notes for a presentations. That information is then sent to the gemini-vision-pro model along with a prompt asking it to review the material based on common best practices. The feedback is presented to the user in the add-on.

The demo uses a zero-shot prompting approach -- no example slides or feedback are provided. Nor has any fine tuning of the base Gemini model been performed.

Exploration and experimentation is encouraged. There are several ways to customize the add-on and improve the quality of the feedback.

• Change the guidelines in the prompt. The current guidelines are generic and cover a subset of recommendations for slides. Change the prompt to fit your own guidelines.
• Switch to few-shot prompting. Providing examples of slides that adhere to your guidelines or violate them may help Gemini provide more relevant feedback.
• Use a fine-tuned model. If you have a large content library with feedback, consider fine-tuning your own model for the task.

These instructions assume some familiarity with Apps Script, Add-ons, and Google Cloud Platform.

1. Create and install an add-on following the quickstart
2. Update the add-on source code & script manifest with the code from the src folder.
3. Enable Vertex AI in the Google Cloud project for the script.
4. Create a service account in the same project and grant it the Vertex AI User role.
5. Create and export a JSON key for the service account.
6. Set the SERVICE_ACCOUNT_KEY script property with the JSON key contents.

Open Google Slides to try the add-on. It should appear in the right hand side panel as an eye icon.