Prompt Hero

Use Prompt Hero to effectively and easily prompt LLMs to do your bidding! Become a Prompt Engineer!

Comment

Inspiration

Prompting is easy, good prompting is hard. While creating another PartyRock app, we realized how hard it was to make really effective prompts. So we decided, let's let the AI "eat it's own dog food" e.g. create an AI app to solve this problem

What it does

Prompt Hero takes a prompt and:

  • Provides suggestions for improving the prompt
  • Provides two prompt rewrites, one simple, one more complex
  • Executes both prompt rewrites so you can see the results of the rewrites
  • Provides a chatbot interface to ask about your prompt, the rewrites or anything prompt related
  • Gives you a new prompt tip every time you load it

Give it a try and check out the video demo

How we built it

We used AWS PartyRock, a new AWS tool that makes it incredibly intuitive to create AI apps.

AWS PartyRock's ability to chain together prompts and prompt output was incredibly useful. It would have been much harder without this.

Challenges we ran into

It was easy to create the skeleton UI for the app; AWS PartyRock enables that. The biggest challenge by far and somewhat ironically was creating the prompt for Prompt Hero. A prompt that creates great prompts, it turns out, is complicated.

We did a ton of research on effective prompting and took copious notes. We then used an LLM to help craft the quite long prompt that Prompt Hero uses. It took a lot of back and forth to get this right.

To finalize the prompt, we used an iterative approach where a Language Model (LLM) was tasked with generating effective prompts through the following steps:

  • Clearly outline the requirements to the LLM and request it to produce a prompt.
  • Review the prompt generated by the LLM, provide feedback, and inquire if it can refine the prompt further for the task.
  • Prompt the LLM to assess the recently created prompt for potential optimizations or adjustments.
  • Iterate through these steps multiple times, allowing the LLM to enhance the prompt with each iteration.

We also found it useful to try a couple of different LLMs when tuning a prompt.

We also had to modify the prompt tips and suggestions to stop suggesting advanced prompting techniques that would not work with our app. A future version of Prompt Hero will incorporate these.

Accomplishments that we're proud of

The prompt itself. It incorporates a lot of information and instructions and works well. Typically you don't need a 782 word, 5700 character prompt but we did for this application.

We find it ironic and somewhat humorous that an App to optimize prompts takes such a large prompt to work well but we also believe this is much of the value of the app, take advantage of the giant prompt we wrote to optimize your own prompt.

What we learned

LLMs can tune prompts for you. Out of the box, they are decent at it but to getting very high qualtiy optimized prompts, you need a complicated prompt in and of itself. Truly this app is a great example of "eating your own dog food" e.g. use the tool to create the solution in the tool

What's next for Prompt Hero

Adding support for various advanced prompting techniques that cannot be shown in the app. Primarily these are techniques that require mutli-step prompts

Built With

Share this project:

Updates