AI · EDUCATION · WEB APPLICATION

IELTS Task 2 Coach

An AI coach that stays with you through the IELTS Task 2 essay, from the first idea to the band score.

APP VIDEO

See how it works.

Question, plan, section coaching, grammar check, self-assessment, band score and progress: the full flow of the app. 2 min 20 sec

ROLE Product, design and full-stack development
PLATFORM Web application
AI 15 structured tools
LANGUAGES English · Turkish

Why did I build it?

The student first plans with the coach, writes the essay section by section with feedback tied to the four official criteria, and then predicts their own band before seeing the score. Coaching is available in English or in a hybrid Turkish mode.

How does a session work?

  1. Unpacking the question: The app generates a question that fits the target band and the student's weakest question type. It highlights key phrases to paraphrase (tap to see alternatives) and shows what the task is really asking.
  2. A 5-minute plan: The student sets a thesis and two body ideas, each with a concrete example. Before writing begins, the plan is rated Weak, Good or Strong.
  3. Writing section by section: Each paragraph has its own guide, a sample sentence, topic-specific vocabulary and tips tagged TR, CC, LR or GRA. The support level can be adjusted from Guided to a 40-minute Exam mode.
  4. Grammar check: Grammar, spelling, punctuation and wording issues are flagged in each paragraph. A hint explains the rule; the correction stays hidden until the student asks for it.
  5. Predict first, then score: The student predicts their band and marks their weakest sentence. The scorer assesses each criterion against the official descriptors, shows the gap from the prediction and suggests one priority fix.
  6. Progress tracking: A current estimate is kept for each criterion, recurring mistakes become focus areas, and the student is moved to a less supported level when ready.

Technical foundation

  • Swappable AI provider: All 15 AI features go through a single structured tool-calling layer. Gemini is the default; Anthropic Claude can be switched on with one environment variable.
  • Output that cannot break the interface: Every response is validated with Zod, retried once with a repair hint if needed, and replaced with safe fallback content otherwise. Broken model output never reaches the student.
  • Hybrid Turkish coaching: In TR mode, explanations and hints are in Turkish, while sample sentences, corrections and vocabulary stay in English, because the essay is written in English.
Next.js 15 React 19 PostgreSQL Drizzle ORM Auth.js v5 Gemini · Claude Zod Tailwind CSS v4 Vitest

Note: The bands the app gives are estimates based on the public IELTS descriptors, not official scores.