One of the major achievements of Intelligent Tutoring Systems is the ability to provide deep diagnosis and intelligent feedback upon user actions in an inner loop of an interactive exercise. In this work, we develop a platform that hosts remote interactive exercises that can be plugged into any browser-based environment. The system supports two feedback modes - a training mode with step-by-step evaluation of student's answer with immediate error-feedback and hint sequences, and a homework or assessment mode, in which a student can enter a complete multi-step solution of a problem, submit it into the system and receive delayed feedback to each made step. Such a detailed analysis of student solutions is provided by the context-aware domain modules used as back-engines.