AI writing checker
Marks Task 1 Academic, Task 1 General Training and Task 2 on all four IELTS criteria. Students see their strengths, what to improve and examples quoted from their own answer. Very short, off topic or memorised answers are capped.
Education case study: PKR 1M+ in its first two months, on a platform built in two.
The founder reported PKR 1M+ in revenue in the first two months after launch, with no ad spend. The platform: an AI writing checker, speaking scores, Easypaisa plans and an admin panel, built in 2 months.
- Client
- IELTS Counsel
- Industry
- Education
- Timeline
- 2 months
- Year
- 2026
IELTS Counsel · Writing Task 2
LiveOverall band
7.0
- Task Response7.0
- Coherence and Cohesion6.5
- Lexical Resource7.0
- Grammatical Range and Accuracy6.5
Read 3 times, middle score usedMentor review available
- Revenue in the first two months, reported by the founder
- PKR 1M+
- Ad spend behind that revenue, as reported
- 0
- From start to launch
- 2 months
- Scoring modes for Writing and Speaking
- 7
( The challenge )
What was getting in the way
IELTS Counsel is an IELTS institute in Islamabad, run by Mubbashar Qayyum. It prepares students for the Academic and General Training tests, in class and online. Writing and Speaking are where students need the most feedback, and they take the longest to mark. Every essay and recording went to an instructor, so students could only practise as often as a teacher had time to mark.
The rest of the business was split across different places. Online students, campus students and students paying in installments were tracked separately. Paid plans needed limits that students could not get around, and paid study material needed protecting. Most students also wanted to pay with Easypaisa, so a card checkout alone would not work.
( Our approach )
How we solved it
We designed the IELTS writing checker so the AI model does the reading and the code does the marking. The model goes through each answer and reports what it finds: grammar and tense errors, vocabulary range, collocations, linking problems, and whether every part of the question was answered. The code then works out each criterion band from that evidence, using fixed thresholds, caps and IELTS style rounding.
Each essay is read three times and the middle result is used, which keeps scores steady from one attempt to the next. We tuned the engine against the institute's own expert marked scripts from Band 4 to Band 9. The first version pushed weak and strong essays towards Band 6, so we corrected that, then ran the new engine alongside the old one before moving students over.
The rest of the platform covers plans and limits, Easypaisa checkout, an admin panel for online and campus students, and mentor review for students who still want a teacher to look at their work.
( What we built )
The pieces that make it work
Delivered in 2 months, with the source code, accounts and documentation handed over in full.
AI speaking scores
Follows the real test across Parts 1, 2 and 3, with a microphone check, question audio and timers. Each recording is transcribed and checked for pace, pauses, hesitation, grammar, vocabulary and pronunciation, and the three parts are combined into one band.
Plans and limits
Free trial, Bronze, Silver and Golden plans control which mock tests, tracks, downloads and AI evaluations a student can use. The server keeps count, and if an evaluation fails the student gets the credit back.
Easypaisa checkout
Students pay from their Easypaisa mobile account. The system checks the payment status and amount, emails a receipt and starts the plan once the payment is confirmed. An old order cannot be used to renew a plan.
Admin panel and analytics
One place for online and campus students, including CNIC or passport records, installments, due dates and balances. Admins also manage tests and blogs, see pending evaluations, and track revenue by source, plan mix and average bands.
Protected study resources
Each paid PDF is marked for the student who downloaded it, with visible and hidden trace marks, so a shared copy can be traced back.
( The outcome )
What changed for IELTS Counsel.
- The business earned PKR 1M+ in revenue on the platform in the first two months after launch, as reported by the founder.
- That revenue came with no ad spend, as reported by the founder.
- The whole platform was built and launched in 2 months.
- Students get a band score and feedback on each criterion as soon as they submit a Writing or Speaking task.
- Mentor review is still available for students who want a teacher's view.
- The institute sells its plans online through Easypaisa, and campus and installment students are managed in the same system.
- Plan limits are checked on the server for every test, download and AI evaluation.
- Admins see revenue, plan mix, pending evaluations and average bands on one dashboard.
- Revenue in the first two months, reported by the founder
- PKR 1M+
- Ad spend behind that revenue, as reported
- 0
- From start to launch
- 2 months
- Scoring modes for Writing and Speaking
- 7
( How it was delivered )
2 months, start to launch
- 1Phase 1
Platform and plans
The student app with Academic and General Training routes, IELTS online practice tests for all four modules, accounts, and plan limits checked on the server.
- 2Phase 2
Payments and operations
Easypaisa checkout, the admin panel, campus and installment students, protected resources and the analytics dashboard.
- 3Phase 3
Writing and speaking scores
The writing checker and the speaking pipeline, from reading the answer or the recording through to criterion bands and feedback.
- 4Phase 4
Tune and roll out
We tuned the scores against expert marked scripts from Band 4 to Band 9, then ran the new engine next to the old one before switching students over.
( Services used )
- AI & Chatbot DevelopmentAssistants that answer from your own material, and automation that removes the queries nobody wants to answer twice.
- Web DevelopmentFrontend, backend and everything between, built from scratch rather than assembled from plugins.
- Systems IntegrationTwo way connections across payments, CRMs, messaging, calendars and databases.
( Built with )
- React
- FastAPI
- OpenAI Whisper
- Namecheap VPS
- Tailwind CSS
- Next.js
- TypeScript
- Python
- OpenAI API
- spaCy
- Firebase
- Firestore
- Redis
- Easypaisa
- Cloudinary
- Docker
Mainstream tools with deep talent pools, so the client is never dependent on us to maintain what we built.
( Read next / Industrial equipment )
A first website for an industrial supplier, built to be found.
The first website for an Islamabad supplier of industrial testing equipment, built in two weeks with 13 pages, a blog and search optimisation from day one.
Want results like these?
Send a brief and you get a written scope, a timeline and a fixed price back within 48 hours. It costs nothing and commits you to nothing.