AppGen: How we built a platform that makes it super easy for people to create AI apps

‍

Introduction

It all started with an initiative by Albert Padin, our CTO, called AI 30x30. The goal was to create 30 new AI apps in 30 days. At first, some thought it was too ambitious, but many were excited about trying something so challenging. There were apps built by developers that gained traction and media’s attention.

- Travel Budol by Tim, which was featured in both local and national news
- Paralegal Pal by Leigh
- Grognak by Edson
- Prompeteer by Michael (Which is also like AppGen but for ChatBot)
- and more…

I’ve noticed a pattern: we use form fields to get AI results. Our backlog is filled with wishlist items, but we have limited developers. Why not open it up to people without coding experience?

I still remember having that idea while waiting for my partner to buy her milk tea. That feeling of excitement and eagerness to go home and get my hands dirty with coding. I built a proof-of-concept overnight and deployed it to Vercel (I’ll discuss my hosting experience with Vercel later). Then, I launched AppGen to the team.

The idea is to let non-coders build their dream AI-powered website without touching a line of code.

‍

‍

‍

‍

Adoption

The initial version of AppGen required users to manually input their fields in a format similar to this:

Ingredients[placeholder="Enter ingredients please"] - textareaCuisine[placeholder="Enter cuisine"] - dropdown (Filipino, Italian, Mexican, Any)te

‍

Behind the scenes, the AI will code the form fields. It was working, but beta users who don’t have coding experience find the format very difficult to use. Rosie saved the day by creating a UI design for seamless app creation, which is the current experience for AppGen users. Ace also played a crucial role in the development.

‍

‍

There is no edit functionality yet, and there are beta users who have requested it. There has been a lot of back and forth, so I decided to develop the edit functionality.

We also explored the categorization of apps for better organization. We allowed AI to categorize the apps based on the details provided by the user. Shoutout to Ferlie for handling that task.

There’s no login required at that time. Anyone can create their own app and share the link with their network and on social media. Even people outside Symph have tried it, and their apps gained virality. In AppGen’s first month, we aced the AI 30x30 challenge, surpassing the goal of having 30 apps in April by creating 1.2K apps.

‍

‍

Challenges

Some apps gain virality, which means they become attractive to bad actors. During that time, I mentioned that no login was required, which made AppGen vulnerable to a Denial-of-Service (DoS) attack. We were engaged in a cat and mouse game with the attacker. I blocked their IP address, but they quickly changed it. The attacker was incredibly determined. Initially, it was the cost of our AI provider that reached the soft limit, causing the app to stop returning any results. To counter this, I added a reCAPTCHA. However, they still managed to penetrate the security measures. So, I decided to add an additional layer of security using Cloudflare, while also implementing app-level rate-limiting using Upstash. It was a week or two of sleepless nights for me as I stood guard, blocking the attacker and protecting the app.

‍

It went okay for a month, but then the attacker evolved. It’s like a virus. Now, I’m experiencing distributed denial-of-service (DDoS) attacks. It’s interesting to note that Vercel does not provide built-in protection against DDoS attacks. I attempted to reach out to their customer support, but unfortunately, I didn’t receive the assistance I wanted. Being an indie developer, it can sometimes be challenging. As a result of this attack, I received multiple emails from my hosting service provider, Vercel, informing me that my system has reached 1 million execution units and 1 million invocations, among other things. After reaching this quota, Vercel becomes very expensive for additional resources. You can watch the video below for more information.

‍

‍

In my case, I migrated the app to Google Cloud Run , which is a part of the Google Cloud Platform, and added Cloud Armor as an extra layer of protection. So far, I haven’t encountered any issues with it. Hopefully, I won’t have to experience that again.

‍

Breakthrough

The most favorited app right now is CREATE RESEARCH TITLE, created by IanThinker (alias). Its video has been viewed in TikTok 2.7M times, hearted by 166.0K users, shared 12.4K times, and bookmarked by 121.8K people.

‍

‍

Another apps created using AppGen are Lesson Planner PH (4As) / Lesson Planner PH (5Es) both are created by Leigh, which went viral on multiple social media platforms. It has also become the most successful product in Symph under the leadership of JP and Roland. You can learn more about the new Lesson Planner PH here.

‍

‍

New Features

Users are no longer limited to text generation. They can now create apps that generate images and apps that allow users to interact with images, for example, by asking the AI what is in the image or asking AI to convert the Image to Text.

‍

‍

AppGen also has the API of each app so that users can integrate the apps to other platforms or to their workflows.

‍

What’s next?

This tool could be beneficial for students who are learning prompt engineering. My plan is to introduce it to schools so that they can be prepared for the upcoming Generative AI wave.

‍

‍

‍

Right now, we have received some feedback that AppGen is expensive for students. We are working on something to make it more affordable for students.

Is the app already successful? Yes, it has been able to achieve the thing I wanted it to, which is enabling non-coders to create AI apps without having to write a single line of code.

Is it done? No, it can still evolve into something bigger, which means it needs a passionate team to work on it.

‍

‍

---

PS. I used Grammar Check AI to review the grammar in this article.

‍

‍

‍