Building A Web App With AI: Part 2

That quote is attributed to Albert Einstein. I can't verify that he said it but it makes sense and it has practical application. In part one of this series, I used ChatGPT to help me create a web application. That web application was functional in that it did what I needed it to do but it had some drawbacks.

1. I took a long time to return results. Even though the ChatGPT prompt was taking place behind the scenes, there was no way for the user to know and it seemed to take forever to respond. I needed a way to display the response in chunks as you see when you are the ChatGPT interface.
2. The response was being displayed to the user in one, long sentence. All the information was there but it was extremely hard to read. I needed the response to be returned in separate lines and paragraphs to make it more readable.

As in part one, I prompted ChatGPT for code that would create a web application but with a minor adjustment, I requested code that would leverage the stream object available in the OpenAI API.

What's An API?

If you're not familiar with an what an API is, it stands for Application Programming Interface. It's a publicly available way of interacting with software in a programatic way. In other words, instead of using the website interface to get information (e.g. clicking buttons and typing information on a web form), you can create a program that will get the same information for you. If you know what actions you want to perform on a website, you can use an API to automate the process.

One of the ways you can interact with OpenAI is a stream. That's a way of interacting with ChatGPT so that the response you get comes in chunks instead of one big piece. That's what I needed and that's why I specified that in my prompt.

Why Is It So Complicated?

This turned out to be much more difficult than I anticipated. As when I created the original website, ChatGPT returned some Python code and some HTML code but it also returned some Javascript. And, as before, it was just enough code for a basic website. I needed something a little more complex. Not too much, but just enough to make it difficult.

I said it once and I'll say it again, AI will give you a good starting point but it won't solve all your problems. No matter how many ways I prompted ChatGPT, I could not find the answer I was looking for. So went back to searching for examples of how interact Python with ChatGPT using a stream.

Each example I found made heavy use of Javascript, which is not a language I'm familiar with. Now I grant you that, once you understand the basics of programming, it's not that difficult to switch from one programming language to another. But the issue is I really did not want to learn another programming language. The whole point of this exercise was to use AI to become more efficient. If I were to take the time to learn another language syntax, that would defeat the purpose of this project.

So I was faced with a dilema: continue down the path I started on or admit that I might need to start over.

Time To Pivot

I had a very Edison-esque moment when I realized that I need to consider another way of reaching my goal. It wasn't that the path I was on would have failed in terms of functionality, it would have failed in the aspect of time.

So I did some more research and I found a tutorial that uses a framework called Streamlit to build websites. Instead of having to manually write HTML, the Streamlit framework allows you to write all your code in Python (which I'm already familiar with and OpenAI already has a supported library in). So I re-wrote my code using Streamlit and, within an hour, I had a website that did exactly what I wanted. It even returned the response in chunks which is exactly what I needed it to do.

Compared to the hours I'd already invested trying to figure out the solution using HTML and Javascript, Streamlit was a much better option for me.

Next Steps

For now, the code still exists in a GitHub repository. The next step in the process is to figure out a way to host it on a web service like AWS or Azure so I can apply CI/CD principles to it. That will be in the next installment.