Public surface for the repos that actually exist: attention tracking, world models, overlay experiments, Learn, and desktop attention services.
This is no longer framed as a single AR product. It is a set of connected systems: sando-tracker, learning-overlay, Learn, and personal-usability.
Four separate codebases, each with a concrete role, rather than one implied headset product.
Monitor-focus service, world-model editor, and room-layout experiments for understanding where attention is directed across a multi-monitor setup.
Monitor focus service can use camera pose, WT901 IMU, or fallback paths
World-model editor serves room and monitor geometry through local APIs
Chord-focus launcher loads survey results and dual-camera defaults
Learning report tooling summarizes monitor prediction logs
Standalone consumer of control-plane APIs with browser and native overlay surfaces.
Runs as HTTP server, X11 overlay, or both
Exposes study-session, proof-capture, activity, and target-run endpoints
Defines a canonical learning schema shared across the surface
Tests cover focus, images, Emacs context, and activity/event reads
Standalone React/Vite personal learning workbench that surfaces Mesh learning APIs on learn.sandolab.xyz.
Today card aggregates streak, recent reading, resume threads, stale concepts, recall, and ontology progress
Library route covers resource catalog, PDF reader, annotations, reader state, and chapter maps
Recall, glossary, ontology, bridge, proofs, sessions, and diagnostics routes expose the learning loop
Ships as a private Tailscale SPA behind Caddy with /api proxied to Mesh
Local desktop helpers and operator-attention services that make the workstation itself part of the loop.
Defines stable provider contracts for screen focus and operator attention
Ships installable user services for focus daemon and operator attention
Includes status tooling for checking the active desktop stack
Keeps the desktop boundary explicit instead of hiding it inside a browser app
The common thread is not "AR." It is attention, context, and surfaced state across desktop, browser, and study workflows.
The tracking work asks where attention is actually directed before deciding what to surface.
The overlay work is a consumer of existing context, not an excuse to pretend the whole stack is a headset product.
Learn turns context, reading state, recall, proofs, and sessions into explicit study workflows rather than leaving them buried in logs.
This page is intentionally narrower than the old pitch.
What is real: monitor-focus experiments, room/world models, X11 overlay code, Learn study routes, and local desktop attention services.
What is not claimed: a single shipped AR headset product, a field deployment system, or a unified commercial suite.
The useful idea here is the interface boundary between attention tracking, surfaced context, and Learn-backed study work. The repos are early, but they are real.
The next work is about deciding which boundaries are worth hardening.
Relevant context for this slice of the work: University of Manitoba psychology and HCI research, software work across consulting, SkipTheDishes, Datomar Labs, and ReLease / Cios, and the current private control-plane stack.
If you want to talk about attention tracking, overlay surfaces, Learn workflows, or adjacent control-plane design, reach out.
[email protected]