WingPal is a lightweight gen-AI App that I created with my colleague Katherine, trying the no-code platform PartyRock. The idea is to help people create their personal elevator pitches in professional settings like job interviews or casual social situations.

While many people might have already thought of using AI to help them create pitches, not everyone is good at prompt engineering. A good “wingman” should be able to know a lot about the person that he is going to pitch. Therefore, the goal of our project is to simplify the process, making it accessible to non-experts in prompting to create pitches tailored to their background and needs.

Our inputs are like breaking down the question of “Tell me about yourself” into smaller pieces that are more digestible:

1. Direct inputs: Basic information like name, pitch setting, audience, time limit, and style preferences.
2. AI-generated questions: The app asks the user five personalized questions to gain deeper insights.

With these inputs, we crafted and iterated an initial prompt to generate tailored elevator pitches.

Testing and prompt refinement were crucial in this project. We tested various input scenarios, including skipping questions or requesting changes, to ensure robustness. We encountered and addressed issues such as:

• Question count inconsistencies
• Redundant questions about known information
• Grammar issues
• Hallucinations (including unrealistic details)

Through testing, we also discovered limitations in certain pitch types, like casual / flirtatious introductions, where the app performed less effectively.

This project provided valuable lessons in prompt engineering. We learned how to:

• Craft clear prompts
• Break down user inputs for better comprehension
• Test with diverse scenarios for robustness

One aspect we truly appreciated about working with large language models is the relative ease of addressing issues through prompt refinement. Most bugs or limitations could be resolved by simply adding clear and succinct instructions to the prompt, a process far more efficient than traditional software development where significant resources are often required to find and implement solutions.

Overall, this project has been an immensely enjoyable journey that has significantly enhanced our ability to use language models with skill and precision.