POL
The problem
In your first months in a new country you know nobody and you cannot read anything. The ordinary tasks are the ones that break you.
Finding someone to fix a laptop, clean a carpet, cut hair, or move a sofa is easy once you have a network and the language. Without either, every small task turns into an afternoon of asking around in group chats and hoping the person who answers is real.
Local classifieds already existed. They were all written in a language the person could not read yet, which made them useless at exactly the moment they were needed most.
Constraints
I was working alone, with no budget, no content and no users. A marketplace with nothing in it gives a first visitor no reason to return, and without visitors nobody posts. Every decision below is shaped by that.
Three decisions
Translate the listings, not just the interface
Translating the shell of an app is the easy half. What people actually need to read is what another person wrote: the description, the terms, what is included. So POL translates listing content into the reader’s language and shows it in place of the original.
Replacing the text rather than sitting beside it was deliberate. A newcomer reading a second language does not want a comparison view; they want the page to make sense on the first read. The cost is that the original becomes invisible, which is a trade I want to revisit below.
Let the categories grow
I did not try to map every category up front. The taxonomy starts small and gains branches as real demand shows up — a category earns its place when people are already asking for it.
This kept the early product honest. An empty category is worse than a missing one: it tells a visitor the marketplace is dead. The structure had to be designed so that adding a branch later is routine rather than a redesign.
Make the first listing easy without making the tenth pointless
The same form has to work for a shop that posts every week and for someone selling from their kitchen for the first time. Those two people fail in opposite directions: one is slowed down by hand-holding, the other is stopped by a blank field.
I split posting into six short steps, one decision per screen, and put a short piece of coaching inside the steps that people get wrong. Write a clear title and an honest description. Bright, square photos, daylight, one item per photo, no screenshots. A shopkeeper already knows this; a first-time seller does not, and finds out only when nobody calls.
Choosing a category first is a commitment, so the first screen says it can be changed later, and the chosen category stays visible as a chip on the next step so nobody loses their place.
Design for the worst five minutes
The hardest moment for someone who does not speak the language is not shopping. It is an accident, a fire, a person on the floor, and a dispatcher on the phone asking questions in Turkish.
Emergency assistance is built for that moment, and every choice in it assumes panic. The national number and a call button sit at the top and stay pinned to the bottom of every screen underneath, so the primary action never scrolls away. Under it are six situations in plain words — medical, police, fire, accident, I am in danger, general help.
Each situation opens a short list of sentences already written in Turkish, with the English underneath so you know what you are about to say. Play it out loud and hold the phone up, or copy it into a message. There is no typing and nothing to compose, because composing is exactly what a frightened person cannot do. If the sentence you need is not there, you can write it in your own language and get simple Turkish back.
Location gets its own card, with a plain-language line saying it stays on the device unless you choose to share it. In a screen that asks people to trust it during an emergency, saying what happens to their data is part of the design, not a legal footnote.
Give people a reason to open the app before they need a marketplace
On-duty pharmacies and daily currency rates are not marketplace features. Other apps already show which pharmacy is open tonight. I built them anyway, because both are urgent, free and used repeatedly, and a marketplace nobody opens cannot sell anything.
They are the way in. Someone arrives at midnight looking for an open pharmacy, gets a walking route and a phone number, and remembers where the app is. Weeks later, when they need a carpet cleaned, they already know where to look.
Where it stands
POL is live on the web and the App Store. It launched recently enough that I have nothing to report but the build itself — no retention curves, no usage numbers, nothing I would be willing to put in front of you as a result.
What I am watching first is whether the utilities do the job I built them for: whether people who arrive for a pharmacy come back for anything else. That single question decides whether the entry-point bet was right.
What’s next
The translation decision needs a way back. Replacing the original text is right for reading, but business names, prices and conditions are exactly where machine translation fails, and right now a reader cannot tell when it has. A small control to show the seller’s original wording would cost almost nothing and would keep the reader in charge of what they trust.
Posting gets a fast lane. Six steps is right for a first listing and slow for a tenth, so repeating a previous ad and starting from a draft are next. I designed for the harder user first, deliberately, and the frequent seller is the other half of that job.
And measurement moves earlier. Building alone made it easy to postpone evidence until after launch; the next round of decisions gets numbers behind it before it ships, not after.