Ship a Weekend AI App in 14 Days

Cap the hours before you choose the app

Your borrowed hours are the real constraint. Madhu Pattarambil built Madram Milsan AI in 14 days using the Claude API, Netlify, vanilla JavaScript, and two serverless functions, small pieces of code that run only when needed, and the build avoided frameworks and build steps. He worked while employed full-time, with coding blocks from 5 AM before work, during lunch, and from 9 PM to midnight.

Madhu Pattarambil brings 25 years of chemical engineering background spanning hazardous-waste management, process safety, and industrial manufacturing. A different career does not stop a build that is small enough to finish.

Write down the hours you can actually give, not the hours you wish you had. If your realistic total is a few hours a day, shrink the scope until it fits. The plan needs a daily cap, a launch date, and a list of features you will not build.

Protect the window like a meeting. Match the work to the block: fragile work in the prework window, small edits and tests in lunch, and integration work in the late window. If a block disappears, do not chase it.

Keep the cap boring. A fixed block is easier to protect than a vague promise to work later. When you miss it, ship less.

Choose a stack that disappears

Keep the stack small enough that you can explain it in a short sentence and debug it without a build step. A framework can feel like progress, but it becomes a tax on borrowed time. The usual mistake is adding one because it looks professional; that is where a weekend project becomes a maintenance project. Pick the lean solo stack that lets you change a page, test an API call, and deploy the same evening.

Keep the data model flat. A few fields, a clear prompt, and a readable response are enough. If you need a database, a spreadsheet, or a design system, stop and ask whether the user job can be smaller.

Use one model API, the service that answers the user's question, and keep the prompt narrow. The app should complete the user's job, not become a chatbot that tries to be everything. If the model call fails, show a useful fallback.

Ship a no-account MVP before adding money

Build the smallest version that completes one finishable user job. If the app plans a trip, let it plan a trip. If it compares options, let it compare a small set of options. A stranger should be able to open the page, use it, and leave with a result in a short pass.

Remove the account. Madram Milsan AI was accessible without an account and without a paywall. A no sign-up app removes the main friction and keeps the test honest. The visitor does not need to trust you with an email before seeing value, and the app can still be useful while remembering nothing.

Make the result visible immediately. The user should see the answer, not a loading state that asks for more information. If the app needs more context, ask for it after an early useful result, not before.

Keep the launch scope narrow. Madram Milsan AI covered 140 destinations, spanned more than 80 countries, and supported 17 languages. Treat that as a later expansion, not a launch requirement. The launch version only needs to prove the core use.

Let affiliate links do the billing

Add revenue after the app works. Madram Milsan AI earned revenue from affiliate commissions with 14 travel partners, including Booking.com, Klook, and Airalo. The simplest model is a partner link: the user clicks it, and the app earns a share of the transaction. At launch, a few relevant links in the result are enough; no checkout, no subscription.

Choose partners that fit the user job. If the app helps someone plan a trip, the links can point to flights, hotels, car rentals, and experiences. Keep the list short. A small partner list is enough to test whether the traffic is real.

Do not build a billing system before you have proof that people use the app. Track the clicks, not the vanity metrics. A small number of useful clicks is more telling than a large number of curious visits. If the no-account MVP gets used, the affiliate links become the revenue experiment.