Overview
CapyPrep is a next-generation exam preparation platform built specifically for Vietnamese students preparing for international exams — IELTS, GMAT, SAT, and academic English assessments. Rather than studying from static workbooks, CapyPrep delivers a personalized, adaptive learning experience that adjusts to each learner's proficiency in real time.
The system uses OpenAI's large language models to grade writing submissions, analyze grammar errors sentence by sentence, and provide instant detailed feedback — something a human teacher cannot do for thousands of students simultaneously.
The platform's capybara mascot creates a friendly, low-pressure learning atmosphere that reduces exam anxiety and keeps students engaged long-term.
The Challenge
Vietnamese students typically struggle with the writing and speaking sections of IELTS/GMAT because quality feedback is hard to access. Traditional prep centers are expensive, have fixed schedules, and turnaround time for graded essays can stretch to several days.
There was also no intelligent system that could track progress across individual skills — vocabulary, grammar, reading comprehension, writing — detect weaknesses, and suggest a personalized improvement roadmap.
Our Solution
Ventra Rocket built CapyPrep on a microservices architecture: Next.js 15 with React 19 for a smooth frontend experience, Python FastAPI handling AI logic, and PostgreSQL storing submission history and learning progress data.
The AI writing grading module uses GPT-4o with specialized prompt engineering, evaluating submissions against official IELTS Band Descriptors: Task Achievement, Coherence & Cohesion, Lexical Resource, and Grammatical Range & Accuracy. Results return in under 10 seconds with paragraph-level annotations.
Key Features
- AI Writing Grader: Scores IELTS Task 1 & Task 2 essays from Band 4.0 to 9.0, with detailed error analysis and specific improvement suggestions for each sentence.
- AI Study Buddy: Conversational assistant that answers grammar questions, explains rules with real examples, and conducts role-play speaking practice scenarios.
- Adaptive Quiz Engine: Question sets that automatically adjust difficulty based on recent performance, keeping students in the optimal learning zone (zone of proximal development).
- Progress Dashboard: Skill-level progress charts, predicted band score based on current improvement rate, and comparison against students with similar goals.
- Vocabulary Builder: Intelligent flashcard system with spaced repetition algorithm for durable retention of academic vocabulary.
Impact & Results
In beta testing with 200 students, CapyPrep recorded a task completion rate 3x higher than traditional methods. Feedback turnaround dropped from 2-3 days to under 10 seconds. Students reported significantly higher confidence in writing sections after 4 weeks of regular use.
When fully launched, the platform is designed to serve 10,000+ concurrent students without increasing operational headcount.
Tech Stack Details
Next.js 15 + React 19 was chosen for Server Components support enabling fast page loads even on slow mobile connections — critical for students in regional cities. Python FastAPI handles compute-heavy AI tasks with high performance and easy horizontal scaling. OpenAI GPT-4o delivers best-in-class grading quality with acceptable latency. PostgreSQL with the pgvector extension stores embeddings for semantic search within the sample essay library.