AI products · Web · Mobile
Anta AI Studio is a small product team in İstanbul. We ship our own apps — the system that reasons, the interfaces people actually use, and the measurement that shows whether it works. One is live on both stores. More are on the way.
01 — What we run
Each one started as a job people were doing by hand. Some are in the stores, some are on the way, and some we only run on ourselves.
Freight in Turkey is traded in WhatsApp groups — hundreds of them, scrolling past faster than a driver can read. The app reads the groups a carrier has joined, pulls each load out of the noise, and turns the stream into one feed filtered by route, with an alert when a matching load appears.
A studio where a brand generates the model, dresses it in its own garments, and gets catalogue images without booking a photoshoot. The image models sit behind a prompt layer of our own, so a brand describes the shot rather than writing the prompt.
One frozen shell, many apps. A new app is a fill — palette, copy, prices, and at least one module of real domain logic. The first ones cover tarot, health, and cooking. Their agent layer returns validated, typed turns, and falls back to a deterministic bank when the model does not.
Photograph the clothes you own, put an outfit together, and see it on a model built from your own appearance — as a photo or a short video — before you get dressed.
An assistant that sits in our own working group: it decides whether it has anything to add before it speaks, researches, remembers, and sends reports as PDFs. And a four-agent engineering loop — planner, coder, reviewer, tester — with a human checkpoint before any patch lands.
02 — How we build
Design, engineering, and the AI work sit in the same team. That is what lets a few people keep more than one product alive at once.
The hard part was never the model. It is the system around it — which tools it may call, when it hands off to a person, where an answer came from, and how accuracy gets measured. We design the silence as carefully as the speech.
Navigation, state, billing, analytics, and the design system are built once and frozen. A new product brings its own domain logic and nothing else, which is how a small team keeps several products moving without several codebases.
Our API layer speaks one message format, so changing model is a configuration value rather than a rewrite. Every swap is re-verified against the same checks before it reaches a user — tool calls, output shape, and the repair path for a malformed answer.
Product events land in a store we run ourselves, from first open to subscription. We read the funnel, not a dashboard someone else decided we should look at.
03 — Stack
Model layer to store listing, the same team carries all of it. The stack is chosen per product, then kept boring.
Picked per product. Replaced when a product outgrows it, not when something newer starts trending.
04 — How we work
Every product starts from something people already do by hand, and do badly — reading a hundred group chats for one load, booking a photoshoot for one dress. If nobody is doing it by hand, we do not build it.
A small version, live on both stores, with billing and analytics in it from the first release. Store review is part of the build, not the end of it.
Funnels, reviews, and regression runs on the agent layer, read every week. A product earns its next sprint with numbers, or it stops getting one.
05 — Contact
Partnerships, a question about how something works, or an idea in the same space. A paragraph is enough.
info@antaaistudio.com