Thread by @JoshSabol

Can ChatGPT build an entire web application? Is that possible?

I promise you haven’t seen ChatGPT do what I’m about to show you

👇👇

Here’s why this was insane,

As of April 2023:

🧑‍💻 I didn’t know how to code
🐍 I’d never written a python script
🥄 I didn’t know how a web-scraper worked
💽 I’d never setup a web service, let alone a CI/CD pipeline
🔌 I’d never built an API
💻 I’d never launched a web app

Less than 1 month later, I present to you:

Outlier.Care 🚀

In April 2023, I set out to build “Levels.fyi for Healthcare”.

After an ill-fated attempt, @shl offered to help

https://twitter.com/shl/status/1645265642965315588?s=20

We started from scratch. Literally, the first prompt was:

`Build levels.fyi for doctors in healthcare`

But that just gave us a generic outline:

Sahil and I agreed to let ChatGPT take the wheel and we’ll follow its lead.

One 2-hour Zoom call later, we had a @reactjs frontend, Express backend, @MongoDB database, and @render web service up and running with full CI/CD via Render's @github integration - BOOM!

I thought about what would entice Doctors to want to visit Outlier.Care, and realized many job postings nowadays include compensation ranges. But how to get jobs data?

There’s no way ChatGPT could build a web-scraper, right?

Wrong 🤯

https://twitter.com/JoshSabol/status/1652168763679072256

Wow, ok. Now I have tons of physician comp data from job postings.

Can I show physicians' compensation distribution by specialty right next to each other? Is that even possible?

I’ve never even heard of a chart like that. There’s no way…

🤯🤯🤯

With an MVP design in hand, @shl volunteered again to help.

We loaded the jobs data into MongoDB, piped it into a vanilla HTML table, and then added a Search bar to filter the job list on the frontend…

after only 36 MINUTES with ChatGPT!

Unfortunately @npmjs bugged out on me, so we had to end the recording, but @shl gave me the confidence to continue with the project on my own.

HUGE, ENORMOUS, GRACIOUS THANK YOU to Sahil for your help!!!

🙏🙏🙏

The next step was to replicate the Chart in the MVP Design that ChatGPT wrote in python, and convert it into javascript.

This was by far the hardest challenge yet. After ~4 days struggling to get it to work,

Behold:

I could see the finish line ahead. Just one feature left to build — a way for physicians to share compensation with their peers.

I had a good sense for what data to collect having just helped 3 doctor friends evaluate their job offers.

~5 days and 1,430 lines of JS later:

The last crucial step was to make the site responsive so that it works on mobile.

By now, ChatGPT taught me the gist of CSS. ChatGPT got me going, but it was faster to manually dial in the styling than copy/pasting dozens of times and waiting for GPT-4 to respond.

Voilà 💁‍♂️

THE END 🎬

I’m humbled and mind-blown how far ChatGPT could get me, a former Product Manager who c̶a̶n̶'̶t̶ couldn’t code 6 weeks ago.

@sama / @OpenAI team — you built the tool of the century!

If you haven’t already, check out Outlier.Care and share it with your doctor friends ❤️

If you’re interested in learning more about how this was built (and want to see the exact prompts that were used), here’s a blog about it:
sabol.io/outlier_care

If you enjoyed this thread - do me a favor and go back up to the first tweet and like and share it.

Follow for some hot takes on how this impacts software development in a few days ✌️

And shout out to @stephsmithio, @ShaanVP, and @thesamparr for providing extra motivation to ship this thing ASAP in their recent pod:

👀 If you enjoyed, RT 👇

https://twitter.com/JoshSabol/status/1669052035751108621