I built a SaaS in a week with Next.js, Supabase, and Stripe — here's what I learned

Alberto Towns — Fri, 20 Mar 2026 19:35:51 +0000

Last week I went from idea to deployed SaaS in about a week of evening sessions. Here's the product, the stack, and what I learned.

The product: RentView — a rental property profit tracker for landlords with 1-10 units. It tracks income, expenses, and net cash flow per property per month. No bank integrations, no tenant management, no bloat. Just "is this property making money?"

The stack:

Next.js 14 (App Router) for the frontend and API routes
Supabase for auth, database (Postgres), and row-level security
Stripe for subscriptions ($9/29/49 per month)
Vercel for hosting
Tailwind CSS for styling
Pure SVG for charts (no recharts or Chart.js)

Key decisions:

Supabase RLS means all security is at the database level — users can only query their own data regardless of what the frontend does
No chart library — SVG bar charts and line charts are ~50 lines of code each and add zero bundle size
Mortgage auto-calculates as a monthly expense so users don't have to log it every month
7-day free trial instead of a free tier — free tiers train users not to pay

What I learned:

Choosing what to build took longer than building it. Analysis paralysis is real.
Supabase's new publishable key format works fine with @supabase/ssr
TypeScript strict mode catches real bugs in Vercel builds that don't show up in local dev
The "minimum viable" in MVP really means minimum — ship the ugly version, not the perfect one

Check it out at https://rentview-three.vercel.appFeedback welcome.

I shipped my first SaaS in a day — here's the stack, the costs, and what I learned

Alberto Towns — Tue, 03 Mar 2026 22:22:50 +0000

I've been in analytics and business strategy for 13 years. I know Python. I understand data. I've made a career out of helping companies make smarter decisions.

But until last week, I'd never built anything for myself. Every side project got stuck in planning mode — researching, overthinking, waiting for the perfect moment that never came.

Last week I tried something different. No planning. No Notion docs. Just sit down and build.

What I built

A CSV analyzer. You upload a CSV file and it instantly gives you:

Column-level statistics (mean, median, std dev, unique values)
Interactive charts (histogram, scatter, bar with percentages)
AI-powered analysis that explains your data in plain English

Everything runs in the browser. Your data never gets uploaded to a server. The AI analysis sends only summary statistics to the backend, not your raw data.

The stack

Next.js — frontend and API routes
Anthropic API (Claude) — powers the AI analysis
Stripe — payment processing
Vercel — hosting (free tier)
Plotly — interactive charts

Total infrastructure cost: roughly $1/month.

Security considerations

Since users upload files and AI processes data, I had to think about:

Prompt injection: CSV cells could contain "ignore previous instructions" — I sanitize all data before it reaches the AI with regex filters for common injection patterns
XSS: React auto-escapes, but I also strip HTML tags and neutralize URLs in cell values
API key exposure: The Anthropic API is called server-side via a Next.js API route. The key never reaches the browser.
File bombs: 50MB file size cap, row/column limits per pricing tier, individual cells truncated to 500 characters
Rate limiting: In-memory rate limiter on the API route — 10 requests per minute per IP

Pricing model

Free: 500 rows, 20 columns, stats and charts only
Single Analysis ($2.99): 50K rows, 50 columns, one AI analysis
Pro ($9/month): 500K rows, 100 columns, unlimited AI analyses

The AI analysis costs me about $0.01-0.03 per request, so margins are strong even at $2.99.

First week results

127 visitors
$0 revenue
2 out of 3 Reddit posts got auto-rejected
0 comments on Hacker News
But real people visited and used the tool

Not glamorous. But it's real.

What I learned

The biggest obstacle wasn't technical. The code was the easy part. The hard part was not closing the laptop. Not switching to something easier. Not telling myself I'd do it tomorrow.

If you're stuck in planning mode — here's what worked for me: remove the gap between deciding and doing. Don't make a task list. Don't set up project management. Just open your editor and build the worst possible version. Ship it before your brain talks you out of it.

You can refine later. You can't refine something that doesn't exist.

Try it

csv-analyzer-green.vercel.app

Building in public from here. Feedback welcome — especially if you try uploading a CSV and something breaks.

DEV Community: Alberto Towns