Shayne WinterAI Systems Lab

HomeAI Systems LabCodeConstellation

Case A7 / offline developer tools

Historical project record

CodeConstellation.

An offline developer tool designed to turn code structure and relationships into a navigable semantic 3D space while keeping detail progressive, camera context stable, and the local scanner authoritative.

  • Directed
  • UAT-reviewed
  • Recorded scan
  • Obsidian direction
Recorded graph: Nontrivial semantic code mapRelationships: Cross-project dependency viewEvidence period: Historical project record

Product problem

Flat file trees hide the relationships that make a codebase understandable.

A useful code map needed to reveal semantic structure without turning the main graph into a wall of labels. It also needed to remain local, preserve the established scanner, keep spatial context during selection, and move detailed source information into secondary surfaces.

Interaction requirements

Orientation and progressive detail were acceptance criteria.

  • Operate offline and keep source analysis local.
  • Cluster by semantic role rather than file location alone.
  • Show relationships and orphans without overwhelming the primary graph.
  • Fit the initial graph to the available view.
  • Preserve camera position and perspective during selection.
  • Move detailed source information into detachable or secondary panels.
  • Keep diagnostic fallback rendering outside the main product experience.
  • Preserve the .NET scanner as the authority while exploring an Obsidian companion.

Architecture & tradeoffs

Semantic scan first; 3D navigation and detail surfaces second.

Conceptual CodeConstellation flow.

Local source scan

The established scanner extracts bounded structural and semantic relationships without sending source to a cloud service.

3D relationship graph

Role-based hierarchy, clusters, orphans, and links create a navigable system view.

Stable camera

Selection preserves spatial orientation rather than forcing the user to rediscover the graph.

Source detail surfaces

Secondary panels carry the dense information that would otherwise clutter the primary view.

Conceptual public reconstruction. Source code, repositories, local paths, bridge details, tokens, and private scan data are excluded.

What Shayne owned

Product direction, interaction requirements, and acceptance review.

Product modelDefined the tool as an offline semantic map rather than a decorative file-tree renderer.
Interaction designRequired progressive detail, camera preservation, fitted initial views, role-based hierarchy, and detachable panels.
Integration directionKept the .NET scanner authoritative and shaped the local bridge and Obsidian continuation path.
UATReviewed graph behavior and rejected fallback or diagnostic presentation as the primary experience.

AI role

Human product direction with AI-assisted implementation.

Agents contributed implementation, analysis, design iteration, and supporting tools. Shayne directed product and interaction requirements, established local/offline constraints, and reviewed behavior through user acceptance testing.

Recorded evidence

A meaningful historical semantic graph.

Semantic object mapThe recorded scan represented a nontrivial application structure.
Relationship graphDependencies and connections were modeled across the codebase.
.NET scan scopeSource projects and supporting smoke tools were included.
Obsidian directionCompanion integration work established a continuation path.

Result & next stage

A substantial developer-tools record with a clear continuation path.

The record supports product direction, interaction design, local/offline requirements, UAT, and a nontrivial semantic scan. The next validation stage is a fresh build and runtime review followed by the planned Obsidian companion work.

Scope and project structure come from the final curated historical project record.
Continue in the AI Systems LabGeometric reasoning and Cortex research