I Built a Production-Ready Indian Tax Calculator in ~1 Hour Using Claude Code — Here’s Why and How

Abhishek Das — Tue, 21 Apr 2026 17:00:17 +0000

Every April I go through the same thing.

Tax declaration deadline shows up. I open my company's internal tax calculator. I enter my deductions — 80C, NPS, health insurance. I look at the tax payable. I change a deduction amount to see how it affects the number.

Nothing changes.

I change it again. Still nothing. The total tax payable just sits there, frozen, not reflecting what I just typed. I refresh. Enter everything again. Same thing.

I don't know if it's a bug or if I'm doing something wrong. I never fully figured it out. I just stopped trusting it.

So I'd go to ClearTax or one of the other popular calculators online. They work, technically. But they're cluttered, the results feel like they're designed to upsell you something, and I'm never quite sure if the numbers are current or if someone updated them for this year's budget rules. Mixed reviews everywhere. No transparency about how the math actually works.

This year I just built my own.

The actual build

I'm a Senior Architect. Nine years of building production systems — apps, mobile, applied AI. I work with these GenAI tools daily, not as a novelty but as part of how I actually work. So when I say I built this with Claude Code, I'm not saying "I typed a prompt and magic happened." I mean I used it the same way I'd use any powerful tool — with intent and some upfront thinking about what I actually needed.

The first thing I did was research. Not code. Research.

I had Claude Code go deep on FY 2026-27 tax rules — both regimes, the new Income Tax Act 2025 that replaces the 1961 Act from April 1st, the Section 87A rebate changes, HRA metro city updates, all of it. I wanted the foundation solid before a single line of code was written. If the inputs are wrong, the calculator is worse than useless — it's confidently wrong.

Only after that did I start building.

The whole thing — calculator, investment optimizer, HRA optimizer, multi-page site, PDF export, SEO setup — took about an hour. Not including buying the domain and DNS configuration, which took its own time.

The tech is React, TypeScript, Tailwind. The calculation engine is pure TypeScript with no external dependencies. All tax data lives in one config file — when next year's budget comes, I update one file and nothing else breaks. I thought about that upfront.

What it actually told me

When I ran my own salary through it, it told me Old Regime was better for my situation.

Which is what I was already leaning toward. I wasn't surprised. But there's a difference between leaning toward something and seeing the exact numbers laid out — this regime costs you this much, the other costs you this much, here's why, here's the slab breakdown, here are the deductions you haven't fully used yet.

It made a fuzzy conviction into a clear decision. That's what I needed.

Why I put it online

I didn't plan to, honestly.

I built it for myself. But when it was done I thought — this is actually useful. Not just to me. Every salaried person in India does this same calculation every April. If my company's calculator has bugs and the public options feel untrustworthy, probably a lot of people are in the same position I was.

So I bought a domain (itrplanner.in), set up Vercel, added Google Analytics, submitted the sitemap to Search Console, and put it live. Free. No login. No ads. All calculations run locally in your browser — your salary numbers never leave your device.

That felt right.

On building this way

People sometimes get weird about the "I built this in an hour" framing. Either they're impressed in a way that feels slightly off, or they're defensive about it.

I'm neither. This is just how development works now, if you choose to use these tools properly.

The thing I'd push back on is the idea that vibe coding means typing a vague prompt and accepting whatever comes out. That's not what happened here. I know how LLMs work. I know where they drift, where they need constraints, how to give them the context they need to produce something accurate and maintainable. That knowledge — what I'd call prompt harness engineering — is what turns a one-hour build into something you'd actually trust with your own tax calculations.

It's a skill. It's learnable. And the people who learn it are going to have a significant edge over those who either ignore these tools or use them lazily.

Try it

itrplanner.in — takes about 2 minutes to enter your numbers and see which regime is better for you. Free, no signup, works on mobile.

If you're a builder and you've been sitting on an idea for a tool that doesn't exist yet — or exists badly — this is a good reminder that the gap between "idea" and "live product" is smaller than it's ever been.

You just have to start with the right prompt. And usually, that's not the code prompt.