AI & language
Chertma
Type ASCII, get correct Uzbek New Latin — ş ç ö ğ. Offline. An open 28-million-row lexicon underneath.
4.89Mword forms, open
Problem
Uzbek New Latin (ş ç ö ğ) is a pain on a normal keyboard. People skip the marks. Search breaks. Keyboards pretend the language is ASCII.
Constraint
Keyboard in, marks out. No install, no account, works with the plane in airplane mode.
What shipped
A converter, a Telegram bot, and a dataset: 4.89M word forms, 23.2M bigrams, a skeleton index for keyboard-indistinguishable forms. CC BY 4.0. The dataset feeds the tool.
The hard part
Keyboard-indistinguishable forms. The skeleton index is the product. The bot is only the door.
Vanilla JSTelegram botHugging Face
More cases
FlareStamina
Free IELTS Academic platform. 144 timed mocks, AI writing feedback, a speaking lab. Teachers run it with a class.
AI Lenta
Uzbek AI news wire. 24 sources, LLM summaries on a schedule, every item cites the original. Nobody in the loop.
Nullsample
Music from code. Zero samples, six stems, same seed = same track. Source public.
Tadam
Wedding photos in one album. One QR, no app, no signup. Used at a real wedding the week it shipped.
KVADRAT
m² calculator for window grilles. Free-text Uzbek and Russian. Offline, PDF out. Built for the shop I work in.
All work