There's a Time and a Place: Unleashing Dynamic Variables in PromptPerfect

Back in the mists of time, PromptPerfect introduced its Prompt-as-a-Service feature, allowing you to serve your prompts via a RESTful API and call them from Python, JavaScript or cURL.

What’s Next for Prompt Engineering? PromptPerfect’s Prompt as a Service!

Deploy prompts and flexible template prompts as REST API services, and integrate them into your applications with just a few clicks

PromptPerfect

Let’s imagine we’re building an AI-powered recipe website. We want users to be able to click a button that suggests a delicious recipe for them. To do that, we can create a prompt and deploy it as a service. We would then incorporate that into our website via Prompt-as-a-Service’s RESTful API.

For example, we could have a prompt that suggested recipes to users:

By calling the prompt, a user could get a new recipe each time:

curl "https://api.promptperfect.jina.ai/RYNiqoT3txBjPoALnV46" \
  -H "x-api-key: token $YOUR_GENERATED_SECRET" \
  -H "content-type: application/json"

tagTemplate prompts

Now, static prompts are fine and all, but we also allow template prompts. This means a user could set options (like ingredients or dietary restrictions) each time they called the prompt:

Which would give the user something like:

curl "https://api.promptperfect.jina.ai/RYNiqoT3txBjPoALnV46" \
  -H "x-api-key: token $YOUR_GENERATED_SECRET" \
  -H "content-type: application/json" \
  --data '{"parameters": {"ingredients":"apples, cinnamon, nutmeg"}}'

tagA time and a place for everything

Now we're taking it up a notch with environmental awareness. Rather than the end-user having to specify where (and when) they are, you can choose to have that information automatically inserted. That means you can create a prompt once, and end-users will get different results depending on their location, date, and time.

Since recipes don’t usually need a specified time and place, let’s consider a new example: Building a website to suggest activities for a user to take part in:

We also need to enable Environment Awareness:

Now, I (in Berlin at noon) get something like this when I call the prompt:

- Visit the Berlin Wall Memorial at Bernauer Straße
- Explore the Museum Island, especially the Pergamon Museum
- Take a walk through the Tiergarten Park
- Enjoy panoramic views from the Berliner Fernsehturm (TV Tower)
- Discover German history at the German Historical Museum
- Stroll along the East Side Gallery
- Check out the Brandenburg Gate at dusk
- Experience contemporary art at the Hamburger Bahnhof museum
- Shop or window-shop at Kurfürstendamm
- Relax in the evening at a traditional German beer garden"

Whereas if my colleague in Tokyo called the prompt, she'd get:

- Visit Tokyo Skytree for panoramic views of the city at night.
- Explore the vibrant streets of Shibuya and see the famous Shibuya Crossing.
- Enjoy the illuminated Tokyo Tower and consider an evening visit.
- Take a stroll through the historic Asakusa district and see Senso-ji Temple.
- Experience the nightlife in Roppongi, known for its clubs and bars.
- Relax at Odaiba Seaside Park and enjoy the night view of Rainbow Bridge.
- Try out some local izakayas (Japanese pubs) for food and drinks.
- Visit a themed café, like a cat café or an owl café, for a unique experience.
- Check out the latest gadgets and tech in Akihabara, the electronics district.
- Attend a traditional Kabuki performance at Kabukiza Theatre in Ginza (if available).
- Indulge in a sushi dinner at one of Tokyo's renowned sushi restaurants.

These variables are great when you're crafting prompts that deliver relevant output based on the user’s time, date, and location. This is ideal for building websites and apps with a global audience.

tagFetching remote data

We can go even further though. With the crawler feature, you can download the main body text of the URL you specify:

This makes it ideal for, say, a summarizer/translator prompt. Here we summarize an article and convert it to a sea shanty:

If I pass the URL for this story on Elon Musk's spaceship exploding, I get this (abridged) output:

(Verse 1)
Oh, gather 'round, me hearties, and a tale I'll tell to thee,
Of SpaceX's Starship, and its flight so brief at sea.
Launched on a morn in November, with power so fierce and grand,
She climbed into the heavens, but 'twas not as Musk had planned.

(Chorus)
Heave ho, lift high, to the stars we aim to fly,
But the rocket's dreams did shatter, and into the sea did die.
Heave ho, lift high, with a fiery tail in sky,
SpaceX's Starship faltered, and we're left to wonder why.

(Verse 2)
The booster and the spacecraft, they parted with a roar,
The engines blazed like comets, as they'd never done before.
The booster met its ending, in a ball of flame it fell,
While Starship soared a moment more, then silence cast its spell.

tagCalling other prompt services

The above example wasn't exactly modular. I had both the summarizer and shantyizer in one prompt. That isn't so useful if I want to shanty all my things in full detail. Luckily, you can also call one prompt from another, allowing for a lot more modularity:

So, if I create a summarizer prompt (named summarizer, which has the sole function of summarizing a web page)…

…I can easily call it from a new shantyizer prompt:

Sending a URL to the shantyizer prompt in turn processes the URL through the summarizer to download and summarize its contents. And of course, the summarizer can be used in any other prompt you like, acting more like a traditional programming language function.

That opens up many more possibilities. Especially when it comes to more complex tasks like cooking a dish. If I’m cooking, for any given recipe I may want one or more of the following steps, but perhaps not all steps for all recipes.

• Download the content of a recipe page
• If it's not in English, translate it (if I'm cooking an authentic Chinese recipe, I want it from a real Chinese website)
• Make it vegetarian (if I'm eating with vegetarian friends)
• Convert it to metric (because reasons)
• Change serving size (depending on how many people I'm eating with)

tagExercise for the reader

As you can imagine, if I'm eating alone (and I'm not a vegetarian), my needs are quite different from when I'm eating with my buddies at the local sea kitten appreciation society. Either way, it means combining several of these "functions".

We can also go multi-modal, combining text and image prompts:

• Create an image generation prompt from the given recipe
• Generate an image of how the food should look

The exercise for you, dear reader, is to build a prompt that performs the above recipe steps (or similar). Share your results on our Discord!

tagMake the magic happen

To get started with magic variables, head to promptperfect.jina.ai and get started. Let us know what you cook up on our Discord!

PromptPerfect - Elevate Your Prompts to Perfection. Prompt Engineering, Optimizing, Debugging and Hosting.

Unlock advanced prompt engineering and prompt optimization for large models such as GPT-4, ChatGPT, Midjourney and Stable Diffusion. Seamlessly deploy your text and image prompts as dedicated services with our free prompt hosting plan. Enhance your large models with superior performance and efficiency.

PromptPerfect