CodeRealm: Adaptive Gamified Learning and Synthetic Tutor

Programming learners often encounter conceptual difficulties that are not adequately addressed by compiler errors or conventional code‐completion tools. On‐demand conversational assistants can provide explanations, but they require learners to recognise that help is needed and formulate a question; unrestricted code generation may also reveal complete solutions before the underlying concept is understood. This paper presents CodeRealm, a proposed browser‐based programming environment that initiates limited assistance from observable editor activity. The system combines inactivity duration, failed executions, repeated error categories, and recent code deletions through an interpretable weighted scoring rule. When the score crosses a configurable threshold and a cooldown condition is satisfied, a backend service constructs a constrained prompt for a locally hosted language model through Ollama. The model is asked to return structured conceptual feedback, a Socratic guiding question, a concept tag, and a Mermaid‐compatible flowchart rather than a complete corrected program. A concept‐mapped practice item then reinforces the same topic, while rule‐based difficulty adjustment uses recent answers and attempts to select the next item. The paper describes the architecture, algorithms, implementation plan, and evaluation protocol. Since controlled empirical testing has not yet been completed, no measured learning gains or system‐performance results are claimed.