Problem
Grading IELTS writing and speaking by hand takes teachers hours, and students wait to find out their band.
What I built
My own startup. I designed and built the product end to end, and went through several versions: a Django monolith, a multi-tenant mock-exam system for multi-branch centres, and a white-label B2B rewrite on Django + Next.js.
Front · Exam app
- Computer-delivered Listening, Reading, Writing and Speaking with real timing.
- Reading highlights, drag-and-drop word banks and multiple-answer questions, like the real exam.
Behind · Servers
- Multi-tenant: every learning centre gets its own branded space with super-admin, manager and student panels.
- An IELTS band-score engine, weighted grading for multiple-answer questions, leaderboards and promo codes.
- Celery + Redis for long AI jobs, S3 for audio, Payme subscriptions, a Telegram bot, Docker and GitHub Actions.
Inside · AI
- Writing is graded by OpenAI and Gemini against the IELTS band descriptors.
- Speaking is recorded, stored on S3 and assessed with Azure Speech.
- ElevenLabs for generated audio.
Results
$600+ in revenue from a learning centre that ran its mock exams on BandBooster for a few months; two centres piloted it in total. It stopped in April 2026. The band-score engine and the AI grading pipeline are the parts I’m proudest of.