Claude Sonnet 4.6 bilan essey baholash, o'zbekcha kelib chiqadigan xatolarni aniq ko'rsatish, haftalik ekzaminator kalibratsiyasi.
Оценка эссе на Claude Sonnet 4.6, подсветка ошибок узбекской L1-интерференции, еженедельная калибровка с экзаменаторами.
The only IELTS coach that understands Uzbek L1 interference. Claude Sonnet 4.6 essay scoring, weekly human-examiner calibration.
Uzbek is SOV and has no articles; English is SVO with the full a / an / the system. When the AI flags an error, it names the 15-code taxonomy below — so students learn the why, not just the fix.
+ 9 more tags (V, D, CU, AUX, WO, AGR, S, PREP, PUN). Full reference card (PDF) →
Practice all four skills with AI-graded feedback, L1 error tagging, a personalised study plan, and a mock exam mode. All inside Telegram.
/exportmydata sends you a JSON within 72 h, /deletemydata permanently deletes after a 30-day grace. Training-corpus use is strictly opt-in and withdrawable any time.Every week we sample 2% of submissions and send them to certified IELTS examiners in Tashkent and Fergana. Their grades become gold labels; the AI is measured against them using Cohen's κ on half-band buckets.
When mean drift exceeds 0.5 bands or κ drops below 0.6, the scoring prompt is re-tuned. You can see the live drift dashboard on the admin side — transparency isn't a marketing line here, it's a cron job.
No credit card, no subscription lock-in. Pay in Stars inside Telegram — the same platform you already use.
Refer a friend, get +3 bonus attempts each. When they go premium, you get +15. Grab your code in the bot →
The things people ask first.
Writing and Speaking rubric scoring runs on Claude Sonnet 4.6 via OpenRouter. Content generation (vocab, reading passages) uses Gemini 2.5 Flash for cost. When you're on a paid tier, every essay gets the Sonnet 4.6 pipeline.
We curated a 15-code taxonomy of the most frequent errors Uzbek learners make in English — article omission (no Uzbek equivalent), case-suffix transfer (Uzbek -da maps to both in and at), SOV/SVO word-order transfer, tense-simplification, false cognates, and 10 more. Every flagged error in your feedback is labelled with its L1 cause so you can fix the habit, not just the sentence. Full reference here.
Yes. Every week ~2% of submissions are randomly sampled into a human-review queue. Certified IELTS examiners grade them; we compare against the AI score, compute Cohen's κ on half-band buckets, and track the drift in a live admin dashboard. If the AI regresses, the scoring prompt gets re-tuned before the next week's sample.
Telegram Stars, one tap inside Telegram — 200 ⭐ (~$3) for a week, 600 ⭐ (~$9) for a month, 4750 ⭐ for a year (55% off). No card entry, no subscription chain you have to cancel. Local card payments via Payme, Click, Uzum Pay are planned for Q3.
All stored on Cloudflare infrastructure outside Uzbekistan; TLS in transit, encrypted at rest. Run /exportmydata to get a JSON of everything we have about you within 72 h. Run /deletemydata and you're permanently erased 30 days later. Using your anonymised work to improve the AI is strictly opt-in and withdrawable any time. Full policy: ielts.rustam.ai/privacy.
Rustam Rakhimov — sole entrepreneur (yakka tartibdagi tadbirkor), Republic of Uzbekistan. Reach me at rakhimovrustam757@gmail.com.
Yes — Band 7.5+ IELTS tutors and ex-official examiners. $1.50–$3.50 per essay depending on tier, remote, flexible hours. Start the application with /examiner_apply in the bot.
Open Telegram, tap the bot, and your first essay is graded in under a minute — with L1-aware feedback that tells you why.
@ielts_rater_bot