AWS released a new playground yesterday (16th November) dubbed PartyRock. PartyRock is an Amazon Bedrock playground that allows you to build small, but powerful generative AI applications based on a natural language prompt.

Consuming the prompt, it will create widgets based on the various outcomes that you wish your application to produce. This is super exciting because it opens the door to incredibly lightweight and useful interfaces that everyone can build - lowering the barrier of entry to GenAI even further.

I got preview access to PartyRock through the AWS Community Builders program and decided to test it out with a fun, and somewhat time-relevant application 😄

🏎 Accelerating to the finish line

Being an avid Formula 1 fan, with the Las Vegas Grand Prix taking place this weekend and AWS re:Invent just round the corner also taking place in Vegas, it only really meant one thing that I could test PartyRock out with.

Enter the F1 Race Guide: Your Ultimate Trackside and Off-Track Planner.

I thought it would be cool to build an application that would consume a single country as a prompt and see what the various AI models in behind the scenes could muster together for some key information if you were ever planning a trip to a F1 race. Some of the key researching points before you go include:

• What are the best seats
• The range of ticket prices
• Best value for money vs. viewing point
• Non-race activities and general tourist/sightseeing activities

So, I thought I’d put PartyRock to the test.

🤖 It all starts with a brief prompt

Like I mentioned above, getting started with PartyRock has an incredibly low barrier to entry - it literally consumes a single natural language prompt to generate your entire application. As with any GenAI application, it’s important to be clear and concise about what you want your application to do and the outcomes you wish to achieve. So I started with the following prompt:

I want the app to be the ultimate Formula 1 companion when it comes to planning and researching attending a race. The app should consume the location the user wishes to research and suggest where the best seats are to sit, the ticket prices, rating of the view point of the seats suggested and a generated image of what the viewpoint might look like. On top of this, it should also generate some general tourist activities that the user can engage with.

Within the minute, PartyRock had the created this nice app on my behalf which included a few widgets based on the prompt.

For the purposes of testing this out, I didn’t fine tune any of the prompts or parameters for each of the widgets - I simply just let the model roll with what it initially generated, but more on the fine tuning later!

♠️ Roll the dice

First of all, let’s test it out with the Grand Prix taking place this weekend - Las Vegas! And like I said, with AWS re:Invent only round the corner, it just feels right.

The app generated the following output as a suggestion for the “Best Seats”:

I absolutely love some of the suggestions including grandstand tickets to general admission, as in some cases depending on the track and what view you’re looking for, general admission can be optimal!

This sounds pretty cool - but with Las Vegas being a new race to the F1 calendar and the track layout changing a few times since its initial announcement, I was curious to cross-reference some of its suggestions with what the confirmed layout was.

Turn 1 grandstands with some hard braking, always great action especially from the start - this checks out! Whilst there is a chicane section of the track, it’s more turns 7-8-9 now with the changes, but close enough and good general advice for some guaranteed action. It also does mention an elusive turn 18, which doesn’t quite exist as we only have 17, so maybe not the best track to start with given it’s brand new and LLM are traditionally slightly outdated.

Let’s take a look at some of the other outputs from the app. For view rating, it doesn’t have enough information from each grandstand location to reliably rate which again, is fair given that it’s a new track and we also haven’t been specific with what grandstand we want to sit in.

Ticket price is a similar story, however it does give some general guidelines and rough estimates which is always welcoming when planning a trip to any race!

Our view image is pretty cool - like an abstract piece of art, almost like a promotional race poster? You’ve got the grandstands, barriers, sweeping flow of traffic - maybe not entirely accurate of what a viewpoint on a particular grandstand would look like, but I think it’s pretty cool nonetheless.

And finally, the tourist suggestions output gave plenty of suggestions for some of the many touristy things to do whilst in Vegas:

All extremely viable options for non-race days, apart from the Bellagio Fountains as it might be quite hard to see them this weekend with all of the race infrastructure in place. Unless you are lucky enough (and have deep enough pockets) for the Bellagio Fountain Club 😎

Overall, pretty cool output - some good suggestions and understandable that some of the information will have been outdated since the release of the track and the changing of the layouts. However, pretty impressed that it was even able to output anything at all given all of those facts above!

🪨 How about something silver?

Let’s give it another go but this time, with a more traditional race that has featured on the F1 calendar over the years with a more defined layout - Silverstone.

For best seats, it gave me a good few solid options including some F1 fan favourites which absolutely checks out (in my opinion!).

If we quickly cross-reference these suggestions with the actual track & grandstand layout, you’ll notice it’s using the specific turn names as that is what they are typically called. Becketts grandstand is in fact, one of the best for Silverstone and the budget option of general admission at Club Corner is accurate as well - although you’ll need to entering the track bright and early to get a good view if that’s the route you’re going!

For view rating, it was a similar story as Vegas - we haven’t been super specific with what grandstand or area we want to sit, so it can’t accurately suggest ratings for the viewpoint.

Ticket prices, again, was similar - not specific enough so can’t be totally accurate, however it did give us general price ranges for a lot of different options including advanced purchasing discount which is pretty cool!

The image it generated you could almost argue looks like it’s the final turn out of the Becketts complex heading down the Hangar straight? Pretty cool piece of generated art though:

And finally, the tourist options for non-race days were as follows:

I would have liked to maybe have seen some suggests for F1 team factory tours if they were available given the context of the app and the fact that there are quite a few teams based not too far from the Silverstone circuit between Brackley and Milton Keynes - but still some solid suggestions!

⚡️ Improving

PartyRock sits on top of AWS Bedrock, so it provides various models to choose from to tailor to your needs and also allows some customisation of how it might react. For example, it will allow you to customise the prompt for each widget that it generates, along with updating the temperature and top-p helping you tailor it to your specific needs.

Conclusion

PartyRock is a pretty awesome, lightweight tool that will allow people to experiment and experience GenAI with a light-hearted touch but also providing insights into what the possibilities with it are.

In this first iteration of PartyRock, it doesn’t have internet access enabled, so that might explain why some of the details weren’t as accurate as they could be. However, I still think given that I deliberately didn’t fine tune anything outside of the initial prompt for testing purposes, it gave pretty solid outcomes that I could actually come away and possibly research further or investigate myself.

The little application that I (or PartyRock I suppose!) crafted is publicly available and you can test it out for yourself! You can even remix it and tweak the prompts yourself to see what other applications you can get it to produce.

Test it out here: https://partyrock.aws/u/pmc-a/l0oODwl7_/F1-Race-Guide%3A-Your-Ultimate-Trackside-and-Off-Track-Planner and please comment/let me know what outcomes and testing you perform with it!

Head on over to AWS PartyRock for more details - it’s free to signup and use for a limited time, so why not give it a try?