With the recent developments in AI, there are an increasing amount of companies and products whose real value add comes from their prompts under the hood. Is it too far (or too cringe) to say we have essentially entered the Prompts as a Service (PaaS) era?
We see teams everyday leverage PromptHub to optimize their prompts for their products. Companies and tools like SmartRecover, Coverdoc.ai and Perfect Chirp are just a few. These companies are of the growing breed that have prompts at the core of their business.
Here are 5 apps you could build quickly whose value comes directly from the effectiveness of their prompts.
App #1: Run Planner
I’m a big runner, and I love using the Nike Run Club app. Fun fact, the Nike Global head coach of running is my old high school cross country coach.
The app has great plans for different types of races that you could be prepping for ( 5k, half-marathon etc), but these can feel sometimes feel a little generic. Maybe that’s by design.
How it works
Regardless, it would be cool to build a personalized running plan app. It would accept user inputs such as their current fitness level,, running experience, goals, available time and maybe even personal preferences like favorite running times and environments.
You could have paid features that allow for hyper-personalization, or maybe even incorporate something like Whisper from OpenAI to have a coach in your ear during runs.
Structuring the prompt
A key aspect of making this idea work is structuring the prompt effectively. The prompt needs to be specific enough to guide the AI but also versatile to adapt to a wide range of user inputs. For instance, the prompt could look something like this:
The prompt is too long to paste fully here. But you can test this prompt in PromptHub and start iterating quickly using our testing tools, like variables, side-by-side comparison, and batch testing. You can test different versions of the prompt. Then, iterate on them based on the results. All versions can be managed in one place. Join our waitlist, or drop me an email to get access!
App #2: CookGenius
Firstly, I know the name isn’t creative, you’ll have to figure out the branding yourself.
I constantly find myself looking at the food in my apartment, trying to think how I can combine chicken breasts, Chobani yogurt, hot sauce, green onions, and whatever else is available to make an edible meal.
Let’s make an app to help streamline this process and generate ideas for dishes based on what I have.
How it works
Users will input their available ingredients, dietary restrictions, cuisine preferences, and cooking skills. We’ll package up these inputs into a prompt and feed it to GPT-4 to generate a few different meal options.
You could monetize via having paid features like meal planning, or integrations with grocery delivery services for items you don’t have.
Structuring the prompt
Effective prompts balance specificity and adaptability. For CookGenius, a prompt might look something like this:
You can iterate on the prompt to refine the outputs, and then build out a front-end in Webflow or Bubble.
The real magic is in the prompt here. The prompt is not just a feature; it's the central offering that provides value to users.
App #3: CareerCoach
A little guidance is always helpful when navigating your career. CareerCoach is an AI-driven career advice platform that provides personalized guidance.
How it works
The guidance will be hyper-personalized based on the user’s current situation, skills, career goals, and interests.
Structuring the prompt
Structuring the prompt effectively is crucial. For CareerCoach, a prompt might look like this:
Take this base prompt, run it through some experiments with different variables, and refine it until the outputs look good. Now you have the core of a small app that you could monetize via ads or paid features.
App #4: GiftGenie
I pride myself on being a good gift-giver, but there is always room for improvement. GiftGenie would recommends personalized gift ideas based on your recipient.
How it works
Our users will input details about the recipient. Their hobbies, interest, the occasion and the budget our user is working with. GiftGenie will output a list of options to inspire our choice.
Structuring the Prompt
The prompt could be something like:
After testing and refining the prompt, you’ll have the core of your application. You could monetize the app via an ad-network.
Idea #5: ThreadMaster
Will Twitter still be a dominate force in social media in the coming years? Hard to tell, but it’s here now, and one of the best ways to increase engagement is by posting threads.
How it works:
ThreadMaster will generate a Twitter thread from longer pieces of content.
Users will input their article, blog post, or report, specify their target audience, desired tone, and the preferred length of the Twitter thread. Then, ThreadMaster breaks down the input into an engaging thread.
Structuring the Prompt:
For ThreadMaster, the prompt will include user inputs and could look something like:
You could monetize this via usage. Users could get 5 free threads a month before needing to pay.
We've developed this prompt in great detail, complete with various examples, and have added it to the PromptHub template library. Join the waitlist here to get exclusive access!
Challenges and opportunities with prompt engineering
Building an app with prompts at the core isn’t as straightforward as it sounds. The Dunning-Kruger effect might have you think that “you just write the prompt and the extremely poweful model gives you the results you are looking for.”
Unfortunately this isn’t the case, or Anthropic wouldn’t be paying $300,000 to hire prompt engineers.
But luckily this is a skill that can be learned. We've written extensively about this (see our article, 10 best practices for prompt engineering with any model), but here are a few quick things to keep in mind.
Understanding your users’s needs
You should be writing your prompts with your users in mind. It can even be helpful to think of a specific user. The more context you can provide in your prompt, the better.
Managing Abiguity
Prompts need to straddle the line between being specific and while giving the model room to think. If you constrain the model too much, you could run into undesirable effects ( the Waluigi Effect). Too vague, and the outputs will be useless.
Iterative Design
Writing prompts is similar to writing code in that it is a very iterative process. It will take many rounds of testing and refining to get a prompt that generates the desired results consistently.
One prompt we built recently went through 45 different iterations, and there are still gains to be made.
Importance of Examples
Examples are king when it comes to getting better outputs.
Having quality and diverse examples can significantly better your outputs.
Leveraging PromptHub
We built PromptHub to help with all things around prompts. Testing, comparing, refining, and collaborationg on prompts. All to help you achieve better outputs and ultimately, happier users.
Wrapping things up
It's an exciting time to be building. You can go from idea to a hyper personalized product in a weekend, without needing to write any code. All you need is to build a solid prompt, and a frontend.
We love seeing what teams using PromptHub are building and we can't wait to see what's next.
