How to build a Generative AI app with no code in three easy steps
Chat GPT and now GPT-4 have taken the world by storm, but how can you leverage them without having to code?
When opening up Chat GPT for the first time, you could feel the magic of a powerful ML model. After some time experimenting, it becomes clear you could build something powerful, but how? Many tutorials involve knowing how to code, which can be intimidating. Today we’ll cover how to build a no code ChatGPT app in 10 minutes. In this guide we’ll be building a pizza ordering app, but you can follow along with any creative app ideas you have.
There are three things you need to build no code apps with chat GPT. The
first is you need to generate responses, the second is capturing user data and third is having a conversation loop to keep the assistant going.
Step 0: Making a Voiceflow Account
To get us started we’ll make an account with a No Code tool called Voiceflow, which lets us quickly build powerful conversational ai assistants. You can sign up for a Voiceflow account at https://creator.voiceflow.com/. Once you’ve made an account, create a new assistant and head over to the settings page to enable the AI steps.
Once you have the AI steps enabled, head back to the canvas to start building your first block with the Response AI step.
Step 1: Generating Responses
Let’s start off with generating responses using ChatGPT. You can find the the Voiceflow Response AI in the AI menu and drag it onto your canvas.
Once you have it open, you can type in a prompt; in our example we’ll be asking ChatGPT to list out some items available on a pizza menu. Click preview and you’ll see what it responds with.
Now we can can connect our new Response AI block with our starting block, and hit the play button to run the assistant. And there we go, a basic chatbot using ChatGPT.
Now that we can generate a response on the canvas, how do we feed user data into ChatGPT to handle a full conversation?
Step 2: Capturing user responses
To make our assistant more powerful, let’s add the ability to process user requests. In the listen section, we can grab the capture block and place it after our AI speak step. We’ll listen for the entire user’s reply and save it to the {last_utterance} variable. Variables let us save results so we can use it later.
Now we can ask the user what kind of pizza they want with the speak step and listen for a response. After the response we can use the last utterance variable and feed it into another AI response step as seen below.
We can the prototype again and have ChatGPT list out the menu and answer a question based on the user response.
Now let’s make sure the conversation doesn’t end by adding a conversation loop.
Step 3: The Conversation Loop
To make our assistant more powerful, we’re going to add a way to keep the conversation going, by adding a conversation loop. Conversation loops let us build fluid experiences using the context from previous responses.
With a conversation loop, you’ll be able to create an application that takes in past context, feeds it into a prompt, and generates a new response.
In Voiceflow this is easy to do, by adding a new talk text block to the canvas.
We can then connect the text block to our capture step and ask the user, “Is there anything else I can help with today?”. This will create a loop where the user will tell us what they want and we can keep the conversation going indefinitely.
Conclusion
In this short tutorial we were able to build a pizza parlour with a couple blocks on the Voiceflow canvas. With these new skills, you can now build powerful ChatGPT assistants in a matter of minutes. If you’re interested in learning how to build a more advanced assistant you can checkout our blog post Prompt Chaining For Conversational AI or our youtube series.
Denys Linkov leads the ML team at Voiceflow. His work focuses on making it easier to create engaging conversational experience. He’s also an active member of the ML community as a writer, mentor and speaker.
