Building WritingPrep: An AI-Powered English Writing Feedback Platform Using Gemini
How I built an AI-powered English writing feedback platform using Gemini to help learners receive faster, more accessible, and personalized writing evaluation.
Many students and professionals in Indonesia struggle to improve their English writing consistently.
Whether preparing for high-stakes exams such as IELTS and TOEFL, applying for international scholarships, writing academic essays, or aiming to improve professional communication, high-quality writing feedback is often expensive, slow, and inaccessible.
I wanted to explore how Generative AI could help solve this problem at scale.
That idea led me to build WritingPrep — an AI-powered English writing feedback platform designed to provide fast, personalized, and accessible writing evaluation using Gemini.
The problem
For many learners in Indonesia, improving writing skills is incredibly difficult because feedback loops are severely limited.
Traditional writing correction often depends on:
- Expensive tutors that charge premium hourly rates,
- Delayed manual review that slows down learning momentum,
- Limited practice opportunities due to a lack of structured resources,
- Generic automated grammar checkers that fix typos but fail to explain deeper structural writing issues.
Students need more than just grammar correction. They need comprehensive structure evaluation, coherence analysis, vocabulary enrichment, and actionable improvement guidance. This gap is especially wide for learners outside of major metropolitan hubs, where access to high-quality English education resources is highly restricted due to location or economic barriers.
Why I used Gemini
I chose Gemini because it provides strong natural language understanding and flexible generative capabilities that fit educational feedback systems well.
Gemini enables WritingPrep to:
- analyze essay structure and coherence,
- identify grammar and vocabulary weaknesses,
- provide contextual writing feedback,
- generate personalized improvement suggestions,
- simulate scoring-style evaluations,
- and support interactive writing refinement workflows.
Rather than functioning as a simple chatbot, the platform is designed to act as an AI-assisted writing mentor.
Core features of WritingPrep
The platform is fully functional at app.writingprep.io
AI-generated tasks
To help learners practice consistently, the platform can automatically generate custom, targeted writing prompts and tasks.
Word-by-word AI grammar correction
Instead of generic feedback, the platform highlights specific grammar mistakes word-by-word. It provides the suggested correction alongside a clear explanation of why the change is needed, ensuring users actually learn from their errors.
Detailed AI feedback and estimated band scores
Users receive comprehensive, structured feedback on their submissions. The evaluation contains a general feedback overview, an analysis of the user's specific strengths, clear areas for improvement, and an estimated band score to help them benchmark their current proficiency.
User-centered design
To help learners simulate real test conditions and manage their writing effectively, the interface is designed with built-in tools including a countdown timer, a real-time word counter and dark/light mode switcher.
Comprehensive task history
Users can easily access their past submissions to track their learning progress over time. The historical dashboard captures the entire iteration process for every task, including the original prompt, the submitted text, the word-by-word correction suggestions, and the detailed AI feedback.
Flexible access model
To ensure broad accessibility while maintaining scalability, the platform supports multiple usage tiers:
- Free daily usage quotas,
- Extended usage through personal API integration,
- Premium access models for power users (currently in development as I explore the right payment gateways for the platform).
This architecture helps reduce operational cost barriers while still allowing broader access for learners.
Authentication and user management
Firebase Authentication is used to manage user access securely and support personalized user experiences.
Cloud-based deployment
The platform is deployed using modern cloud infrastructure with Vercel and Firestore to support scalable and responsive usage.
Technical and operational challenges
One of the biggest lessons during development was realizing that building AI products involves much more than prompt engineering.
Real-world AI systems require attention to:
- API security,
- abuse prevention,
- authentication flow,
- rate limiting,
- infrastructure scaling,
- and cost management.
During development, I encountered unauthorized API usage caused by exposed credentials. This forced me to redesign several parts of the system architecture and rethink how AI services should be protected and scaled responsibly.
That experience significantly improved my understanding of operational AI engineering beyond model integration itself.
What I learned
Building WritingPrep taught me that meaningful AI applications should focus on reducing real-world friction.
The goal is not simply to “use AI.”
The goal is to make valuable educational support more accessible, responsive, and scalable for people who may not otherwise have consistent access to high-quality feedback.
Generative AI becomes most impactful when it expands access to learning opportunities.
Future plans
I plan to continue improving WritingPrep by adding:
- deeper writing analytics,
- learning progress tracking,
- adaptive feedback systems,
- multilingual support,
- and more personalized AI learning workflows.
My long-term vision is to build an accessible AI-assisted learning platform that helps English learners improve communication skills more effectively regardless of location or economic background.
Final thoughts
WritingPrep started as a small experiment exploring how Generative AI could improve writing education accessibility.
Over time, it evolved into a real platform involving AI integration, cloud infrastructure, authentication systems, usage management, and operational security considerations.
This project continues to shape how I think about building practical and responsible AI-powered educational tools.
Try out the live platform today at app.writingprep.io.