Shayne WinterAI Systems Lab

HomeAI Systems LabCodeConstellation

Active developer-tools project

Active development · prior runtime exercised

Make an unfamiliar codebase easier to see, navigate, and explain.

CodeConstellation turns local code structure and relationships into an explorable 3D map, while keeping detailed source information in supporting panels. It is designed for understanding complex and AI-assisted codebases without sending source to a cloud service.

  • Product defined
  • Implemented
  • Runtime exercised
  • Active refinement
Domain: Offline developer toolsRole: Product direction, interaction requirements, UAT, acceptanceStatus: Active local source with prior runtime evidence
For a hiring managerThis project demonstrates the ability to turn “help me understand this code” into a clear product experience: decide what information belongs in the overview, preserve the user’s orientation, keep analysis local, and verify that the interface remains useful on a nontrivial codebase.

The product problem

File trees show where code lives, not how the system fits together.

Large and AI-assisted codebases can be difficult to understand because the important relationships span folders, projects, types, and dependencies. CodeConstellation provides a system-level view first, then lets the user inspect the relevant source detail without crowding the primary map.

What the product enables

Explore structure without losing context.

Semantic overview

Groups code by role and relationship so the user can see major components, dependencies, and disconnected areas.

Stable navigation

Preserves camera position and perspective during selection so exploration does not become repeated reorientation.

Progressive detail

Keeps dense source information in secondary panels and supports an Obsidian companion for notes and continued analysis.

My ownership

Product direction, interaction design, and acceptance review.

Product modelDefined the tool as a semantic system map rather than a decorative file-tree renderer.
Interaction requirementsRequired progressive detail, stable camera context, useful initial framing, and clear hierarchy.
Privacy boundaryKept source analysis local and treated the local scanner as the authority for what the visualization may claim.
User acceptanceReviewed real graph behavior and rejected diagnostic fallback views as the primary product experience.

Current evidence

The project is active, with prior native and Obsidian runtime exercise.

The current local source has been accepted into the active workspace and development continues. Earlier runtime acceptance exercised a nontrivial local codebase, a visible native 3D canvas, selection without camera loss, bounded source details, and note creation through the Obsidian companion.

Current maturity is active product development with prior runtime evidence. The next checkpoint is a fresh current-source build and package review focused on product behavior, interaction quality, and repeatable acceptance.

Transferable capabilities

What this project demonstrates.

  • Developer-tool product direction
  • Complex-system visualization
  • Semantic code analysis
  • Progressive disclosure
  • Interaction design
  • User acceptance testing
  • Offline / local operation
  • .NET / WPF
  • Three.js / WebView2
  • Obsidian integration
Next AI Systems Lab case studyPegasus Exchange