Lately I’ve been craving projects that are quick to build, fun to use, and strong enough to include in a portfolio. So I gave myself a challenge:
Can I build something useful in 20 minutes — and ship it the same day?
Spoiler: I could… kind of.
🧠 The 20-Minute Method
The idea is simple:
- Pick a tiny, delightful idea
- Use AI tools to scaffold it fast
- Deploy it, polish it, share it
- Then decide if it’s worth going deeper
I originally started building an Instagram Reel script generator for ecommerce brands. It had potential, but the AI logic wasn’t strong enough out of the box, and I found myself over-engineering it. For a 20-minute project, that’s death.
So I pivoted.
💡 Enter: The Bullshit Detector
The premise is simple:
Paste a piece of text — a LinkedIn post, a startup pitch, or a marketing blurb — and the tool tells you:
✅ Not Bullshit
🚨 100% Bullshit
…or somewhere in between.
It started as a joke. But I quickly realized: this actually scratches a real itch. We’re all tired of buzzwords and fluff. I wanted to create a little radar that helps people check their own writing—or call out corporate nonsense with receipts.
🎛️ Rules, Not Models
This project is a completely different beast from my earlier Hot Dog / Not Hot Dog classifier, which used Teachable Machine and real machine learning to train a visual model.
Instead of “learning,” the Bullshit Detector uses a custom set of rules (written by me) to score the text. That means it’s fast to build and 100% subjective. But that’s also the point: it’s fun, fast, and hopefully funny.
🧰 The Stack
Here’s what I used to build it:
- Lovable – to scaffold the basic UI and get a working version live fast
- ChatGPT – to help shape the prompt and output format
- DeepSeek – for the language model that scores the actual BS factor
- GitHub – for version control
- Vercel – to deploy it in seconds
🛠️ Still Debugging…
I got the tool live fast, but I’m still working on the logic.
The main issue? It’s too nice.
Even when you paste in marketing word salad, it rarely triggers a high enough BS score. So I’m tweaking the logic and language model instructions to push it toward more honest — and occasionally brutal — results.
🗺️ Want to Build Your Own? Here’s the Quick Path:
- Pick a small, punchy idea — bonus points if it makes people laugh or nod
- Use Lovable to create a fast front-end
- Write a good prompt — this is the secret sauce
- Export the project to GitHub
- Deploy in one click via Vercel
- Add your own logic (or hook it into a model like DeepSeek or OpenAI)
- Test, tweak, laugh, share
- Don’t overthink it. Ship it.
🚀 Final Thoughts
This project wasn’t about being perfect — it was about shipping fast.
I didn’t overpolish it. I didn’t spend two weeks debating font choices. I just made a thing, pushed it live, and started using it.
Sometimes the best thing you can do is build in public, laugh a little, and improve as you go.
So if you’ve got an idea like this in your notes app — build it. The bar for usefulness is lower than you think, and the payoff of seeing something you made live on the internet is still the best.
Link to my [work in progress] bs detector: https://truth-or-tripe-test.vercel.app/
