Inspiration
The idea for Unbabel Mind came from a simple observation: learning English is hard, and most platforms don’t offer the flexibility, empathy, or personalized experience that learners need.
"What if learners could talk to an AI that feels like a patient tutor and a friendly partner?" That vision inspired us to build a smarter way to learn English—conversational, contextual, and always available.
What it does
Unbabel Mind helps users:
- [x] Practice English conversation in real-time
- [x] Get grammar corrections with explanations
- [x] Learn vocabulary in natural contexts
- [x] Track learning history
- [ ] Speak with AI (coming soon!)
How we built it
We broke the project into key components, each responsible for a part of the learner’s experience:
| 🧩 Component | ⚙️ Stack | 📋 Description |
|---|---|---|
| Frontend | React + Tailwind CSS | Responsive, interactive UI with chat and dashboard |
| Backend | Node.js + Express | Handles requests, sessions, and API communication |
| AI Integration | OpenAI GPT API | Grammar correction, explanations, dialogue |
| Database | PostgreSQL | User data, vocabulary tracking, chat logs |
| Spaced Repetition | Custom SRS Logic | Vocabulary reviews based on memory decay theory |
Challenges we ran into
Every meaningful project comes with challenges. Some of the biggest were:
Handling User Input Variability: Learners make mistakes—that’s expected. But AI sometimes misunderstood their intent or overcorrected valid expressions. I had to iteratively improve prompt engineering and add fallback logic. Latency & Cost: API calls are expensive and sometimes slow. I implemented local caching for repeat queries and adjusted the system to reuse certain AI responses efficiently. Tone and Empathy: Language learners need encouragement, not robotic feedback. I experimented with different AI personas and found that a friendly, supportive tone was most effective for learning. Security: Since users are inputting personal sentences, I needed to ensure safe data handling and filter inappropriate content before it reached the AI.
Accomplishments that we're proud of
Created a working MVP in just a few weeks Helped early testers feel more confident writing in English Developed an intuitive UI that made users feel at ease Built an AI that doesn’t just correct, but teaches
One tester even said: "It’s the first time I’ve felt like I was chatting, not just learning."
What we learned
Prompt engineering is as much art as science Teaching with AI requires empathy, not just data Small design choices (like saying “Great try!”) can boost learner confidence Language learning is not about perfection—it’s about progress
What's next for Unbabel Mind
We have big plans for Unbabel Mind:
Speech input and pronunciation feedback Roleplay mode for real-life scenarios (e.g. ordering coffee, job interviews) Teacher dashboard for classroom use Multilingual support so more users can benefit Mobile app to make it more accessible globally
Ultimately, we aim to create the most human, helpful, and accessible English learning tool—powered by AI, but always rooted in empathy.
Built With
- next.js
- node.js
- postgresql
Log in or sign up for Devpost to join the conversation.