A full-stack web application in which users can practice coding questions under contest style constraints that enables real-time code execution and provides AI-driven feedback on time and space complexity and code improvement suggestions at the end of test.
Built for students, competitive programmers, and developers who want instant execution, contest-style practice, and intelligent code analysis.
-
We wanted to do direct deloyment of our server through render and client through vercel and so we ran the render deployment on free tier and due to the storage size limitation the deployment broke!!, teh large size is due to intial model loading, even though it loads only once, still its not doable through free tier :(
-
But we will try deploy it next year on paid version or some other way which mostly requires a paid plan !
- Real-time code execution via Piston API.
- Contest-style timers and resizable editor panels for an interactive workspace.
- AI-powered time/space complexity analysis using a fine-tuned CodeT5 model deployed on Hugging Face.
- Integrated with Google Gemini GenAI API for detailed code improvement suggestions.
- Database-backed question management and user authentication with scalable backend design.
- Frontend: React (Vite)
- Backend: FastAPI, SQLAlchemy
- AI Models: Fine-tuned CodeT5 (Hugging Face), Google Gemini GenAI
- Execution Engine: Piston API
- Database: Initially used a local SQLite database, later migrated to PostgreSQL hosted on Render (free tier)
- Processed and cleaned dataset of ~5000 coding questions
- Built and evaluated baseline models (Logistic Regression, MLP) achieving 63% accuracy
- Improved performance to 81.8% using CodeT5 fine-tuning for complexity analysis
- Published the fine-tuned model on Hugging Face for global access
- Students preparing for coding interviews
- Competitive programmers practicing under timed conditions
- Developers exploring AI-assisted coding feedback
Q: How does CodeSmart analyze time & space complexity?
A: Code is passed to a fine-tuned CodeT5 model on Hugging Face, enhanced with Gemini GenAI for detailed feedback.
Q: What languages are supported?
A: All languages supported by Piston API, but for simplcity the code editor allows only python, C++, java.