🚀 Inspiration
Learning to code isn't just about getting the right answer—it's about developing the ability to think through problems. However, most existing AI coding assistants prioritize generating solutions over fostering critical thinking. We noticed that students often become dependent on AI, skipping the reasoning process entirely.
We wanted to change that.
Our goal was to build an AI-powered coding mentor that supports students without replacing their thinking. Instead of solving problems for them, our system understands when a learner is struggling, predicts when intervention is needed, and provides personalized hints that encourage independent problem-solving.
💡 What it does
Our project is an intelligent VS Code extension that acts as a real-time AI learning coach.
It continuously analyzes coding behavior such as:
- Typing patterns
- Compile and runtime errors
- Time spent on tasks
- Debugging attempts
- Hint requests
- Problem-solving progress
Using these signals, our AI predicts when a student is stuck and delivers adaptive, context-aware hints instead of complete solutions.
A coach dashboard provides educators with live insights into student progress, allowing them to monitor learning, identify struggling students, and provide timely intervention.
🛠️ How we built it
We built our solution using a modern AI-powered architecture:
- Frontend: React + TypeScript
- Extension: VS Code Extension API
- Backend: FastAPI (Python)
- AI Models: Gemini API + Machine Learning recommendation engine
- ML Pipeline: Behavioral feature extraction, session analysis, intervention prediction, and personalized hint generation
- Data: Coding session telemetry including errors, typing activity, execution history, and interaction patterns
Our recommendation engine combines behavioral analytics with LLM-generated educational guidance to provide personalized support tailored to each learner's progress.
⚡ Challenges we ran into
One of our biggest challenges was balancing AI assistance with genuine learning. We had to ensure that the system guided students without simply giving away answers.
Other challenges included:
- Designing meaningful behavioral features from coding sessions
- Detecting when a student genuinely needed help
- Generating educational hints instead of solutions
- Synchronizing real-time communication between the VS Code extension, backend, and instructor dashboard
- Optimizing AI response quality while maintaining low latency
📚 What we learned
This project taught us that building educational AI is fundamentally different from building a traditional coding assistant.
We gained hands-on experience with:
- AI-assisted software engineering
- Machine learning for behavioral analytics
- LLM prompt engineering
- Real-time event processing
- VS Code extension development
- FastAPI backend architecture
- Designing AI systems that enhance human learning instead of replacing it
🌍 What's next
Our vision is to create the next generation of AI-powered learning environments.
Future plans include:
- Personalized learning profiles
- Adaptive difficulty prediction
- Instructor analytics powered by AI
- Multi-language programming support
- LMS integration
- Privacy-preserving on-device intelligence
- Predictive learning interventions based on long-term student progress
We believe AI should empower learners—not think for them.
Built With
- ai
- analytics
- api
- artificial
- code
- engineering
- extension
- fastapi
- gemini
- generative
- git
- github
- intelligence
- javascript
- learning
- llm
- machine
- prompt
- python
- react
- rest
- sqlite
- typescript
- vs
- websockets
