I have a confession. I have a B.Tech in Computer Science. I understand programming concepts. I can read code and mostly understand what it does.
But I did not write the code for my website in the traditional sense. I built the entire thing using Google AI Studio. Prompt by prompt. Component by component.
And it works. It is live. Real people use it.
Let me tell you how.
What I Built
A personal finance calculator platform for Indians called MonuMoney.in. It has 36 financial calculators (8 fully functional right now, 28 being built), a blog, SEO setup, and the whole thing runs on Next.js 15 with Tailwind CSS deployed on Vercel.
The calculators do real-time EMI calculations, SIP return projections, income tax computation based on Indian tax slabs, retirement corpus planning, and more.
This is not a toy project. The calculations need to be accurate because people will make real financial decisions based on the numbers.
What Vibe Coding Actually Means
There is a lot of hype around "vibe coding" right now, and most of it misses the point. Let me explain what it actually looks like in practice.
Vibe coding is not "tell AI to build me an app." That does not work for anything beyond a demo.
Vibe coding is more like pair programming where your AI partner does the typing and you do the thinking. You need to know what you want. You need to understand the architecture. You need to be able to read the output and know if it is right or wrong.
Here is my actual workflow:
Step 1: I decide what I need. For example, "I need an EMI calculator page with SEO metadata, structured data, and a client-side interactive component."
Step 2: I write a detailed prompt that specifies exactly what I want. Not "build me an EMI calculator" but a prompt that describes the inputs, the formula, the output display, the styling classes, the file structure, the metadata format, everything.
Step 3: Google AI Studio generates the code. I read through it. Sometimes it is 90% right and I ask it to fix the remaining issues. Sometimes it completely misunderstands what I wanted and I rephrase.
Step 4: I paste the final code into my project, test it locally, and push to GitHub. Vercel deploys it automatically.
This cycle takes anywhere from 20 minutes for a simple component to 3 or 4 hours for a complex calculator page.
Why Google AI Studio and Not Cursor or Copilot
No particular reason honestly. I started with Google AI Studio because it was free and I was already comfortable with the Google ecosystem. It just stuck.
Could I do the same thing with Cursor, Copilot, or Claude? Probably yes. The tool matters less than the skill of writing clear, specific prompts.
Things That Work Well With This Approach
Building UI components is where vibe coding really shines. Tailwind CSS components, page layouts, responsive grids, card designs. These are things where the AI has seen millions of examples and can generate solid output consistently.
SEO boilerplate is another winner. Metadata objects, structured data JSON-LD, sitemap generation, robots.txt configuration. This stuff is repetitive and well-documented. AI handles it perfectly every time.
Standard financial formulas also work great. EMI formula, compound interest, SIP returns. These formulas are well known and the AI implements them correctly.
Things That Do Not Work Well
Anything involving complex state management across multiple components gets messy. The AI generates code that works in isolation but does not always integrate cleanly with existing components. You end up spending time debugging state conflicts.
Design consistency is another challenge. If you generate each component in a separate prompt, they might use slightly different spacing, colors, or patterns. I learned to always include my existing design tokens and class names in every prompt.
Edge cases in financial calculations are the scariest part. The AI might implement the basic EMI formula correctly but miss edge cases like what happens when the interest rate is zero, or when the tenure is less than one month. You absolutely need to test every calculator with extreme values.
The Skill That Actually Matters
The most important skill in vibe coding is not prompt engineering. It is knowing what to ask for.
If you do not understand how a Next.js App Router works, you cannot write a good prompt for it. If you do not know what structured data is, you cannot ask the AI to implement it. If you do not understand the EMI formula, you cannot verify the AI's output.
My B.Tech CS background helps a lot here. Not because I write code, but because I can read it. I understand file structures, component hierarchies, client vs server rendering, and basic algorithmic logic. When the AI generates something wrong, I can usually tell and explain what needs to change.
Vibe coding does not eliminate the need for technical knowledge. It eliminates the need for typing speed and syntax memorization. Those are very different things.
The Financial Accuracy Problem
This deserves its own section because it is the most important consideration for a project like this.
When you build a calculator that people use for financial decisions, accuracy is not optional. If my EMI calculator shows a wrong number and someone budgets based on it, that is a real problem.
My approach:
Every formula is cross-verified against at least two established sources. I check against known correct values from bank websites or government portals.
I test with known inputs and expected outputs. For example, I know that a Rs 10 lakh loan at 8% for 10 years should give an EMI of Rs 12,133. If my calculator shows anything different, something is wrong.
I always include a disclaimer that calculations are for educational and estimation purposes only. Not because I am not confident in the accuracy, but because financial tools should always carry this disclaimer.
Would I Recommend This Approach
For building tool-based websites and utility apps, absolutely yes. Vibe coding with an AI assistant is faster than traditional coding for someone at my skill level. Probably 3 to 4 times faster.
For building complex applications with lots of interconnected state, authentication, databases, and real-time features, I would be more careful. At some point, the complexity of prompts exceeds the complexity of just writing the code yourself.
For my use case, a collection of independent calculators on a static site, vibe coding is perfect. Each calculator is self-contained. Each page follows the same pattern. The architecture is simple and repetitive. This is exactly where AI-assisted coding performs best.
The site is live at https://monumoney.in if you want to see the end result.
If you have questions about the workflow or specific prompts I used, drop them in the comments. Happy to share more details.
Top comments (0)