Intelligent Competitive Programming Companion – ICPC

Training young people in programming is crucial for addressing 21st-century challenges. This project focuses on creating an AI assistant that supports learning programming, particularly for competitive programming. Using reinforcement learning, the assistant will offer hints and exercises to improve learning outcomes without revealing solutions. The project aims to scale up tutoring by automating feedback, reducing dependence on human tutors for careless errors.

Programming competitions provide a rich dataset of problems that are easy to grade and involve real-life scenarios. The AI will focus on personalized feedback using machine learning techniques, such as code embeddings, to analyze and improve code through iterative refinements. Open-source models like Mistral and LLaMa 3.1 will be used to reduce costs and ensure data privacy.

Key project goals include creating a graphical representation of programming concepts, using reinforcement learning to guide learners efficiently, and developing adversarial methods to generate challenging test cases. A randomized controlled trial (RCT) will validate the AI’s effectiveness in real learning environments. The project also aims to identify and analyze student errors, using this information to generate useful hints while avoiding giving away solutions directly.