What I've built, the real problem each one solves, and where it runs — the open-source libraries other engineers depend on, and the high-load systems behind real products.
Published packages, built to be depended on by other engineers.
One Python library that connects your app to all of Uzbekistan's payment providers through a single, consistent API.
The problemEvery Uzbek product re-builds the same five fragile payment integrations from scratch — the exact code most likely to leak money.
A FastAPI library that adds passkey / WebAuthn login — the “sign in with Face ID or fingerprint” standard — without taking over the rest of your authentication.
The problemWebAuthn is the strongest login we have and the easiest to implement subtly wrong, leaving a lock that looks closed but isn't.
A FastAPI library for trusted-device and session management — know every device on an account, and cut any one of them off instantly.
The problemLogin proves who you are once; the real risk lives in the sessions and devices afterward — which most apps never track and can't cleanly revoke.
Infrastructure and applied-AI work — mostly private by nature, serving real users every day.
The main marketplace of the Soff education ecosystem — ready-made academic materials, 3D designs, website templates and freelance services.
The problemKeep search, checkout and browsing instant while the catalogue grows past a million items and thousands of people shop at the same time.
An AI assistant that generates presentations, course papers, reports and tests in seconds — on the web and inside Telegram.
The problemTurn a one-line prompt into a finished, correctly-formatted academic document in seconds — from either a browser or a Telegram bot.
The internal control room for the Soff ecosystem — users, sellers, sales pipelines and analytics for the marketplace and the AI platform, all in one place.
The problemA marketplace and an AI product generate operational chaos with no single source of truth for who's who and what's selling.
Production AI automation built into large business products — turning slow, manual, repetitive processes into LLM pipelines that run unattended, stay observable and fail safely.
The problemMost “AI features” are demos that break on real business data, volume and edge cases — with no answer for what happens when the model is wrong.
A prototype AI agent for Biotact Deutschland GmbH that takes a product brief and returns a full quarter of marketing: a costed content plan, example assets, and targeting — not just a wall of copy.
The problemMost “AI marketing” tools generate a paragraph on demand; nobody hands the team a whole quarter's plan they can actually schedule and cost.