built end to end
I'm Roma. I build apps that live inside Telegram: someone taps a button in a bot and lands straight in the interface — no install, no sign-up, no passwords.
I take the whole project: screens, server, database, Stars payments, deployment to a VPS and support after launch. No need to hire a frontend dev, a backend dev and someone who knows Docker separately. I build with a model alongside me, so timelines are shorter than usual and the price is lower.
services
Four areas that usually come as a set. You can take one, but most projects arrive whole — from the first screen to the server that serves it.
A real app inside the messenger. It opens from a button, inherits the user's theme and feels like part of Telegram rather than a website in a frame.
A bot as the entry point to a mini app and as a product of its own: flows, broadcasts, payments, an admin panel and scheduled notifications.
A server that survives a rush of users and never loses an order. Typed API, honest migrations, validation of everything arriving from outside.
A project doesn't end at the code. I package it in Docker, put it on a server, close it behind HTTPS and leave instructions anyone can follow.
work
Two projects live in Telegram right now. Both built entirely by me: interface, server, database, bot, payments and deployment.
Prompt store
Catalogue
Mini App · paid in Stars
A storefront of ready-made prompts for Claude, ChatGPT, DeepSeek and Alice. You open the shop from the bot, pick a prompt, pay in Stars — and the text opens right away, with no waiting and no messaging a seller.
Stack
Podsekay
Fishing companion
04:20
in-game time
real 21:15
Recent sessions
Mini App · for Russian Fishing 4
A companion for Russian Fishing 4 players. It measures how profitable a session is while it happens: how much silver per hour a spot brings, how much went on bait, and whether the spot is worth staying on. The phone sits next to the keyboard and finally earns its place.
Stack
approach
A mini app is not a website opened inside Telegram. It's a separate kind of product with its own rules: no address bar, no tabs, no patience for loading. If the first screen stalls, there is no second screen — the person closes the window and goes back to the chat.
So I don't start with design. I start by asking which single action the person opens the app for. Everything else is built around it, and extra screens get cut before anyone writes them. The project ends up smaller, cheaper and shipping sooner.
A model speeds the build up. The routine — standard screens, plumbing, migrations, tests — it writes faster than I do, and that saves the project weeks. The decisions stay mine: what we build, how it works inside, where the security boundaries are and what actually ships. Code I did not understand does not go into the project.
After that it's ordinary engineering with no magic: typed code, validation of everything from outside, database migrations instead of manual edits, tests on the money paths, and a compose file that brings the whole thing up with one command on any machine.
TypeScript or Python across the whole project. Logic doesn't drift between services, and nothing has to be written twice.
No secret and nothing to be shy about: the model writes the routine, I keep the architecture, the security and the review. That is why a mini app costs less and ships sooner than one assembled by hand from scratch.
Environment variables, initData verified by HMAC with the bot token, rate limiting and a strict CSP. The token never reaches the repository.
README, deployment guide and a list of commands. The project can be passed to another developer without a half-day walkthrough.
After launch I stay on the project: fixing what real users surface and adding features as it grows.
contact
Describe in a couple of sentences what you want to build — I'll tell you whether it works inside Telegram, how long it takes and what it costs. If it isn't my kind of task, I'll say so instead of wasting your time.
I reply within a day. If it's urgent, say so and we'll move faster. Talking costs nothing: you can simply ask whether a mini app makes sense for your case at all.