HeapKeep: Exploring Novel Codebases Through an RPG Interface

Despite academic studies suggesting that software visualization can support developers in daily tasks, visualization tools remain underutilized in industry. This suggests a disconnect between the design of software visualization tools and developer engagement. Video games offer an alternative interaction model that is immersive, intuitive and broadly accessible. This work introduces HeapKeep, a game-like spatial representation of code, augmented with large language model (LLM) assistance. We evaluate how such a tool may impact developer onboarding for unfamiliar codebases. HeapKeep was developed in Unreal Engine 5 to represent a codebase as a navigable 3D dungeon where rooms correspond to classes and code health metrics shape the level layout. An LLM-powered non-player character (NPC) accompanies the player, answering code-related queries. A small user study compared HeapKeep to traditional text-based code exploration. Thematic analysis of participant interviews suggests that LLM assistance may accelerate orientation in a novel codebase. However, participants tended to rely more on the 2D in-game map than on the 3D environment, raising questions about the utility of spatial representation versus abstract visualizations.