Vibe Coding: How I Used AI to Code an Entire Parenting App in 1 Month!

Mar 29, 2025

Hey guys, it’s Ritz here, and I’m really excited to share how I “Vibe Coded” an entire Parenting AI app in 1 month.

For those wondering, “Vibe Coding” is a term coined by Andrej Karpathy in February 2025, referring to the collaborative process of working with AI to code projects beyond your current skill level — essentially vibing with the AI to create something amazing together.

From Desperate Parent to App Founder

It all started when I became a parent. Like any new parent, I faced numerous challenges, especially when it came to finding reliable information. I often asked ChatGPT for help with certain problems, but the results were inconsistent. This was particularly frustrating with sleep training — we tried various methods to get our kids to sleep, but nothing seemed to work until we eventually hired a sleep trainer who helped us achieve our goal.

That’s the thing with parenting information — it’s scattered everywhere and often conflicting. I thought to myself: what if I could create an AI app that correlated the best information from across the internet, using open sources, and turned that knowledge into a chatbot? Essentially, a better version of ChatGPT, but specifically for parents.

How RAG Powers my ‘ParentGPT’ App

I started with what I knew best — Python. I’m quite proficient with it and coded most of the app’s backend using it. For the knowledge base, I used a RAG (Retrieval-Augmented Generation) engine called LlamaIndex, which gave excellent results in terms of accuracy and referencing source material.

The development process was significantly enhanced by AI coding assistants. I initially tried Cursor, which has some great features for pair programming with AI but had a limitation — it couldn’t search through my files automatically. I had to explicitly tell it which file to reference, which became a roadblock.

That’s when I discovered Windsurf Codeium, which solved this problem beautifully. I used Windsurf about 90% of the time, and when I hit limitations, I switched to Cline, which allowed me to use just about any model. Both Cursor and Windsurf Codeium have their strengths, but for my workflow, Windsurf’s file search capabilities were crucial.

How I Coded a Full Flutter App with Zero Experience

Once I had this great capability, the next challenge was figuring out how to release it to the world. I decided to try Flutter, despite having absolutely no experience with Dart or the Flutter framework. I was completely oblivious to it, but the cross-platform capabilities were too appealing to ignore — I could code for Android, iOS, and web simultaneously.

There was a significant struggle along the way with the Flutter development. However, if I had gone the manual route, the struggle would have been even greater. I likely would have needed to hire a developer, which would have been very expensive. AI really helped me express my ideas through code and generate an app that’s quite functional today, with a professional look and feel.

My background in graphic design and aesthetics gave me an advantage here. I could direct the AI to create a particular look and feel that aligned with my vision for the app.

The Challenges of Vibe Coding

When it came to Vibe Coding, there were several significant challenges. The biggest issue occurred when the AI got stuck in loops, thinking it had found a solution when it hadn’t. It would go around in circles, stuck in what I call a “global minima,” and it took time to overcome these situations.

Going back to first principles helped — I’d explain to the AI how to approach the problem step by step. Start with the backend, move to the frontend, figure out the data flow, and so on. This methodical approach worked well for specific situations.

I found that models like Claude 3.5 performed exceptionally well for coding tasks, while Windsurf’s implementation of GPT-4o, o1 and o3 mini struggled significantly. Interestingly, Cline didn’t seem to have the same implementation issues with these models.

With the release of Claude 3.7 Sonnet, I noticed a substantial improvement — the number of iterations needed decreased dramatically. The model is clearly smarter, allowing me to reach solutions much faster with less struggle.

How I Beat AI’s Knowledge Cutoff Problem

Another challenge was that AI models weren’t trained on the latest repositories. If a codebase was updated recently, the AI wouldn’t have knowledge of it. I solved this through a mixture of approaches:

Copying source code back and forth from the repos
Using MCP servers to ingest new GitHub repositories
Leveraging Windsurf’s web search functionality to access the latest documentation online

These techniques were crucial when integrating services like:

Firebase for user signups and authentication
RevenueCat for in-app payments
Facebook SDK for ad tracking
Creating APIs and connecting the backend
Deploying the backend to AWS

Personalized, Science-Backed AI for Sleep-Deprived Parents

While Vibe Coding did much of the heavy lifting, I was solely responsible for the direction and management of what the app would look like, what it would do, and how it would solve real problems for parents.

The resulting app, ARIA, is designed specifically for parents and caregivers. Users can create profiles with their child’s information — date of birth, specific problems, medications, allergies, behavioral development milestones, and more. ARIA takes this information and, based on expert science-backed knowledge, provides answers that are context-aware and backed by experts.

One of my favorite features is the memory function, which allows ARIA to remember your conversations and provide increasingly personalized knowledge over time.

ARIA is now available on both Google Play Store and Apple App Store for parents everywhere.

Vibe Marketing: My Next AI Frontier

Now that ARIA is deployed on both app stores, I still have improvements I want to make. However, before diving deeper into development, I need to tackle marketing — another behemoth challenge. I’m very new to marketing, so I’m using AI to help me figure things out.

Perhaps there’s a “vibe marketing” approach waiting to be discovered — using AI to scrape leads, send personalized messages, and generate targeted ads. In my last newsletter, I mentioned a technology called Animatic, which I’m using to generate full end-to-end videos from a single prompt. I’m currently testing whether I can create 20 videos a week while I sleep to drive revenue and bring users to ARIA.

If you’re interested in any of these technologies or want to learn more about my journey building ARIA, feel free to contact me on LinkedIn, WhatsApp. Until next time, see you then!

Stay connected with news and updates!

Join our newsletter to receive the latest news and updates from our team.
Don't worry, your information will not be shared.

We hate SPAM. We will never sell your information, for any reason.