How to use EngProgress

EngProgress is an AI examiner for IELTS and for the Uzbek Multilevel (CEFR) exam. It writes practice for you, marks it against the official criteria, and tells you the one thing standing between the band you got and the next half band up. All four IELTS skills are live — Writing, Reading, Listening and Speaking — and the Multilevel exam has its own Reading and Writing papers in their own format rather than IELTS with the labels changed.

It starts wherever you are. If you have never sat the exam and would score a 4, the first tasks are written at a 4 and the coaching assumes nothing; if you are pushing from 7.5 to a 9, they are written there instead. There is no entry level to clear before the platform is useful to you, and no ceiling once you are good.

It is built for someone preparing on their own. Nothing waits on a teacher: you ask for a task, it is generated and served immediately, and the report comes back in the same sitting. You are not left alone with it either — you can ask the tutor for help in the middle of a task, and it will answer without giving the answer away. Education centers run the same platform with teachers, groups and assigned homework on top of it — that has its own guide, linked at the foot of this page.

The part worth understanding before anything else is the marking. Every competitor can put a band on an essay; the number is only worth having if it survives exam day. So the grader is calibrated to sit slightly low: when your work falls between two bands it gives you the lower one and names precisely what the higher one was missing. A 6.5 here is meant to be a real 6.5 in the exam hall, and being told you are not there yet is far cheaper than finding out in July.

There is no question bank here and no set of tests to work through. Every task is written the moment you ask for it, and the level it is written at comes from you — your measured band for a reading paper, the level you pick for a listening test, a topic and question shape you have not been given before for an essay. Two learners practising on the same day sit different papers, and you never sit the same one twice.

Ask for a task and one is written on the spot. The generator draws a topic family from the fourteen it rotates through — environment, education, technology, health, work, society, government, globalisation, crime, media, culture, transport, tourism — and, for Task 2, one of the six question shapes the real exam uses: opinion, discussion, problem–solution, two-part, advantages versus disadvantages, and positive or negative development. Before serving it, it checks every prompt you have already been given and every essay you have already written, so the same question never comes round twice. Academic Task 1 goes a step further and draws the chart itself — and the grader is handed that same chart, so it marks the figure you actually saw rather than somebody's description of it.

This is the skill that reads your level off your own results. Before a word is written, the generator looks up your measured reading band — the one your previous attempts produced — and falls back to your target band, and then to a sensible default, if you are new. A full test is built around that number rather than flat at it: passage 1 lands a band below you, passage 2 at your level and passage 3 a band above, which is how the real paper ramps. One authentic Cambridge question layout is chosen for the whole test so the three passages cohere, the order of the question blocks inside each passage is then shuffled, and each passage is given a different subject and angle so you are not reading three variations on one theme. Every question is finally checked back against the passage it came from, and anything that cannot be confirmed there is thrown away rather than served to you.

There are two ways in. The shared library holds practices that are already recorded, so they open and play at once. Or make your own: choose a level from L1 to L5 and the engine writes an original script and then performs it as multi-voice audio — a complete four-part, forty-question test in about two and a half minutes, or a single ten-question practice in about two. The level does far more than change the vocabulary; it drives the delivery. A lecture is voiced at roughly 115 words a minute at the easy end and about 140 at the hard end. A phone conversation moves from unhurried turn-taking to a fast native pace with barely a gap between speakers, and a student discussion goes from lively to genuinely overlapping. On a quick practice the question format is withheld until the announcer states it, so you cannot rehearse one type and hope it comes up.

The examiner is not reading from a fixed list — and the list is not frozen at the model's training cutoff either. The pool of themes is refreshed once a day by a grounded search and shared across the platform, so it cannot quietly go stale, which is the worst way for content to age because nobody notices. From there the two modes pull in deliberately opposite directions. Exam practice keeps timeless, exam-shaped themes — your home town, food, routines, and the durable Part 3 questions about how society changes — because questions built from this week's headlines would be less exam-authentic, not more. Free conversation gets the current topics themselves, because discussing what is actually going on is the whole point of that practice. If a search ever fails, the built-in pool takes over: a lesson never waits on one and never fails because of one.

The generator was built against the structure of the official practice-book format: the parts, the question types, the layouts, the pacing. Listening comes out as four parts and forty questions — form, note, table and sentence completion in Parts 1 and 4; map and plan labelling, matching and multiple choice in Part 2; discussion multiple choice, choose-TWO and flow-charts in Part 3 — with multi-voice audio and the standard narrator framing. Reading comes out as three academic passages of rising difficulty carrying matching headings, matching features, True/False/Not Given, Yes/No/Not Given, note completion, sentence endings, pick-TWO and summary completion, with the group-level word-limit instructions printed the way the books print them. Writing covers Task 1 reports and every modern Task 2 category, including the newer “outweigh” and positive-or-negative-development phrasings. Listening and Reading are then converted on the standard forty-question raw-score table.

There are four of them, and they are deliberately different things. The Writing tutor and the Reading tutor sit with you DURING the task: ask what belongs in this paragraph, or which two words in this sentence to compare against the passage, and you get a straight answer in the moment. Both are hard-blocked from doing the work for you until you submit — the Writing tutor will not write a sentence of your essay and the Reading tutor will not tell you whether Q7 is True, no matter how you ask. They teach the move on a different example instead, and the full explanations unlock the second you hand the work in, which is what keeps the band yours. The Speaking tutor works the other way round, because speech is live: it reacts, corrects and teaches on every turn while you are talking. The study coach is not attached to a task at all — it sees your bands, your weakest skill and how many days you have left, and tells you what to do with the time.

The Multilevel papers are generated on demand like everything else, but against the CEFR descriptors rather than the IELTS band descriptors, because they are a different exam with a different mark scheme. Reading is built as five parts and thirty-five questions in the shapes the real paper uses; Writing is built as all three tasks. A result comes back as a CEFR level rather than a band, which is what the certificate actually reports.

There is a separate guide for education centers

Teachers, groups, student logins, assigned homework, Telegram notifications, attendance, reports and finance — all of it is covered in its own guide.