Building a Digital Sommelier on Top of ChatGPT and the OpenAI API

Yes, thank you so much for having me. My name is Jan. I'm a lead developer at EddDwye and today I want to show you how to build a digital AI-powered sommelier through Bracel.ai SDK. And initially it was planned that this talk is part of the Remote Day on Tuesday. So when I recorded the first version a couple of weeks back, I talked about the frantic pace at which things are changing in the AI space. And I guess the latest development really proved that point. And I thought, is my talk still relevant? Because a couple of weeks back, we received custom versions of JetGPT. Maybe you have tried it out. You can build your own versions of JetGPT for any kind of topic. And guess what? There are plenty of sommeliers among them and they are actually pretty good. So I thought, great. What does it mean for my talk? And then Sam Altman was ousted as CEO at OpenAI. Most of its employees went on strike and I thought, okay, how long will this API still be around? But since then things calmed down a bit. Altman is back as CEO at OpenAI. And yeah, I think when I thought about it, the underlying message of my talk is still true. That is that AI, especially generative AI, will have a profound effect on how we work. The good thing is, we as developers can shape that how. We are the ones building and designing the interfaces that facilitate between users and large language models. And that gives us a really good position in this transition to using generative AI tools. The best thing is, it's actually really simple to get started. It doesn't take more than five minutes. So let's pick a challenge and see how far we get.

So maybe you guessed it from the topic. I am really into wine and I like pairing food and wine, but there are literally millions of possibilities from various regions and countries, France, Spain, Italy, you name it. So finding the right pairing can be really challenging. But we can build a really simple, hopefully smart, digital sommelier that helps us with finding that perfect pairing.

So let's do that in the next five minutes with the Resell AI SDK and the OpenAI API. So we'll start with a really simple UI. I'm not lying if I tell you that this took the longest to put together. So we have two columns, there is a menu on the left, so the food, and there is a wine list on the right to pair with the food. And on the bottom, you have a small form, a text area where you can add new dishes to the menu.

If you look at the code, it's really simple. We make use of the use chat hook provided by the Resell AI SDK, and it gives us a set of helper functions to interact with large language models. So it does all the heavy lifting, there's not much we need to do here. And we have the form itself, which uses some of these helper functions to send the data off to an API road. That is also actually quite simple. On the server, we have three things, we initiate a client to interact with the OpenAI API, we specify the model. In this case, it's GPT 3.5 TORGO. We provide our OpenAI API key. Unfortunately, it's not free. And that's it.

We read the form data from the request, so the dish in this case we want to pair a wine with, and we send it off to the OpenAI API. What we get back is a streaming response, or a response that we turn into a readable screen that we send back to the client. And here, there's not a lot more to it. Again, we use chat hook to render this streaming response on the screen. And that gives us that nice effect that looks like the AI is actually typing the response. Probably if you use chat GPT, you've seen it before. And that's really all there is to it. Maybe a hundred lines of code. But I think it's a good showcase of what you can do really quickly with open source tools. But let's see if that actually works. So I prepared a little example.

So this is our actual sommelier. Let's start with fresh oysters on ice and let's see what we get. What kind of wine to pair with. So it suggests us a champagne. It's a really classic pairing. So I guess a good start.

Let's do something for Christmas. Maybe a roast duck with red wine sauce next. And it pairs with a pinot noir. So it could be from France, a red wine or from California. Who knows? And it probably would go really well. It's not very specific, but yeah, why not?

And let's finish off with a dessert. Maybe Mousse au chocolat is nice for this time of the year. And we get a port wine. I guess there are some people from Portugal here. So probably you know that. It's a really classic pairing. Sweet wine with a sweet dessert. And that's really all there is to it.

I don't know if it replaces actual sommeliers anytime soon. But what I wanted to do with this talk is basically to show you it is really simple to get started. It's literally not even a hundred lines of code. We use freely available tools, open source tools. I think there are even some contributors here. So I use Next.js, the Brazilian SDK and for the pretty UI, ShadeCN. So I can highly recommend those. But the gist of this is it's really easy to get started. You don't need to be an AI expert. Just try it out. Pick a challenge. Deploy something to RSL or Netlify or whatever and see how it goes. I think what we should keep in mind if you work on that in some more serious applications, think about how users will interact with generative AI or these tools. Think about the responsible uses and then have fun with it.

And yeah, that's really all there is to my talk. If you liked it, please also check out And Why. We are a small design and technology studio from Munich. We have an office here in Berlin as well. And we do next-to-AI development, 3D and immersive experiences on the web.