I Let AI Build My Apps in Replit - And

 Things Got Interesting 🤯

If my last blog was about discovering vibe coding, this one is about going deeper.

This time, I used Replit - and I built three things in one session that I genuinely didn't expect to pull off.

First, What Is Replit?

Replit is a browser-based coding platform - no installation, no setup, no "it works on my machine" pain. You open a browser, start a project and write or run code right there.

But what makes it different now is the AI agent built right into it. You describe what you want, and it doesn't just write code , it thinks through the problem, plans the steps, installs the dependencies, builds the app and runs it. You can literally watch it work. That "AI is thinking" panel? It's oddly satisfying to watch.

And the best part - you can see the full code it generated. Every file, every line, fully open.


               

Project 1: Landmark Map Explorer 🗺️

I started simple. I asked Replit's AI agent:

"Make an app that displays a map of local landmarks based on the current viewable area of the map. Use Wikipedia to get the landmarks."

15 messages and 42 actions later - the app was live.

I could pan around the map, and it would automatically pull real Wikipedia landmark data for whatever area was visible. No hardcoded places. No fake pins. Actual landmarks, dynamically loaded.

That alone already felt like something worth showing people.

Project 2: Human Detection System 🎯

Then I wanted to push it further. I asked it to build a Python AI web app that detects humans in real time using a webcam. Full prompt, proper requirements - YOLOv8 or MobileNet SSD, bounding boxes, confidence scores, FPS display, detection logs, screenshot capture.

48 actions. 8 minutes. Running app.

The dashboard it produced looked like this:


Live. Real numbers. Updating in real time.

I recorded it with OBS Studio because honestly, it deserved to be on video.

And the controls panel? Fully interactive:

  • Confidence threshold slider - I set mine to 45%
  • Crowd alert trigger - set to 6 people by default
  • Toggle switches for face blurring, sound alert, and entry/exit line
  • Action buttons - Save Screenshot, Reset Heatmap, Download Report
  • A live Detection Log updating in real time

This wasn't a static UI. Every slider and toggle was wired to the backend - change the confidence threshold and the model immediately adjusts what it considers a valid detection. Hit "Download Report" and you get a full timestamped log of the entire session.



I recorded it with OBS Studio because honestly, it deserved to be on video.

But here's the honest part - this kind of AI model needs real processing power to run smoothly. On a regular laptop, the FPS stays low and it can feel sluggish. This isn't a Replit limitation or a code problem - it's just the reality of running computer vision models. For a truly smooth experience you'd need a machine with a dedicated GPU.

That said , the fact that it built and ran at all, in a browser, with no local setup? Still impressive to me as someone just starting out.

Project 3: The Upgraded Version 🔥

I wasn't done. I went back and said - make it better. And it upgraded the entire system with:

  • Face/head blurring for privacy (Gaussian blur over the head region)
  • Heatmap showing where people have been detected most over time
  • Entry/Exit counting using centroid tracking across a virtual line
  • Crowd threshold alerts - animated red banner when too many people are detected
  • Auto-screenshot when the crowd limit is exceeded
  • Downloadable detection report with full session logs
  • Four view modes - Detection, Heatmap, Privacy, Entry-Exit tabs

This went from a beginner portfolio project to something that genuinely looks like smart city surveillance software.

And yes - the webcam did give me a bit of trouble at first (camera access in browser environments can be finicky). But once it was running, watching it detect, count and log in real time was something else.

What Made This Different

With Lovable (which I wrote about in my last blog), I was building frontend apps and changing themes.

With Replit, I was running Python backends, AI models, OpenCV pipelines - things that traditionally take days to set up - in a single afternoon.

The AI didn't just write code. It planned. It debugged. It upgraded. I could see every decision it made.

What's Next For Me

I'm not stopping here. I've already planned my next steps:

  • Learning AI prompting properly - how to write prompts that get better results
  • Exploring Coursera for structured AI and ML courses
  • Keeping building and documenting everything here

My Honest Take 💬

AI is not something to be scared of or dismissive about.

In a few hours, I built a landmark explorer that pulls live Wikipedia data, and a surveillance-grade human detection system with heatmaps, face blurring and crowd alerts - using nothing but a browser and good prompts.

I'm still a student. I'm still learning. But the gap between "idea" and "working app" is shrinking fast.

Try something new with AI. Break things. See what happens.

You might surprise yourself. 🚀

Blog #6 in my NetWithHiru series - documenting everything I learn as I go. Videos recorded using OBS Studio. Projects built on Replit.



Comments

Post a Comment

Popular posts from this blog

Nervous, Excited, and Proud — My pfSense Configuration Journey

Image
  My pf Sense Configuration Journey    .🖉 When my lecturer told us we would be setting up a firewall using pfSense, I honestly had no idea what I was getting into. This is the story of how I went from zero experience to successfully logging into a pfSense dashboard and everything that went wrong (and right) along the wa y. Starting Fresh: VMware for the First Time The first thing I had to do was set up a virtual machine environment. I had heard about VMware before, but I had never actually used it. My go-to had always been VirtualBox so even before I touched pfSense, I was already stepping into unfamiliar territory. Installing VMware was a new experience on its own. The interface looked different from what I was used to, and I had to take some time to get comfortable with how it worked. It was a small hurdle, but it reminded me that in this module, almost everything would be new. Installing pfSense: More Steps Than I Expected Once VMware was ready, it was time to...
Read more

NetWithHiru-Blog-Series-01

Image
  My First Networking Project: Connecting Accounts and Delivery Departments Introduction This project focuses on the design and implementation of an inter-departmental network using Cisco Packet Tracer. The network connects two departments, Accounts and Delivery, ensuring efficient communication through proper IP addressing and routing configurations. The project demonstrates fundamental networking concepts such as subnetting, device configuration, and connectivity testing in a simulated environment. Objectives To design a simple network connecting the Accounts and Delivery departments To configure IP addresses, subnet masks, and default gateways using the 192.168.40.0 network To implement routing between two different networks using a router To connect all devices using appropriate network cables To verify network connectivity between departments using the ping command To gain practical experience in using Cisco Packet Tracer for network s...
Read more

Ethical Hacking Journey

Image
Blog #3 - I Installed Kali Linux Before My Lecturer Even Mentioned It One big reason I got into this path is Wenura Mendis sir and his streaming channel Vibernet Sreaming . Watching his content opened my eyes to the world of ethical hacking and cybersecurity in a way that felt real and practical not just theory. It inspired me to stop watching and start doing. So when I enrolled in the Cisco Networking Academy Ethical Hacking course, I didn't want to just follow along in class. I wanted to get my hands dirty before anyone told me to. That's where this whole Kali Linux journey began. "I didn't wait to be taught. I broke things, fixed them, broke them again - and learned more than any classroom could have given me in a single week." If you've been following my blog series, you already know I don't wait around. The moment I enrolled in the Cisco Networking Academy Ethical Hacking course, I knew I needed Kali Linux on my machine. Not next week. Not when ...
Read more