Productivity Systems / 2025-2026
Eisenhower AI / Matrix Systems
A productivity system for prioritization, scheduling, and future AI-assisted task triage.
- Role
- Desktop app engineering, workflow design, AI product direction
- Status
- Active direction
Evidence
What I owned and how it can be checked
My contribution
- Designed and built the original Python/PyQt desktop application, including quadrant interactions, persistence, settings, import/export, statistics, and Windows packaging.
- Re-architected the product as a TypeScript monorepo with a React/Tauri desktop shell, shared task rules, and a Markdown vault storage package.
- Implemented optional local AI classification through an OpenAI-compatible LM Studio endpoint, with review-first controls and dedicated evaluation scripts.
Verified evidence
- Public application sourceThe public Python/PyQt repository documents SQLite storage, drag-and-drop, statistics, JSON/CSV transfer, system-tray behavior, and standalone Windows packaging.
- Local-first rewriteThe active private rewrite contains React/Tauri desktop code, shared core and Markdown-store packages, vault-safety tests, and Node-based verification commands.
- Versioned Windows betasPrivate beta folders contain dated setup executables, MSI packages, portable executables, and checksum files for verified release iterations.
- AI evaluation pathClassifier, coaching, provider, settings, and startup modules are paired with tests, while offline and local-model evaluation scripts remain separate from the product UI.
Desktop productivity interface organized around urgency and importance.
Key decision
How much authority should AI have over a user's priorities?
Choice
Let the model propose a quadrant and rationale, then keep the user in control of confirmation and final placement.
Why
Priority is a commitment, not only a text-classification problem. A review step makes the suggestion explainable, reversible, and compatible with a local-first workflow.
Tradeoff
Confirmation adds a small interaction cost, but avoids silent task movement and preserves trust when a model is uncertain.
Premise
A serious local-first productivity surface
The project starts from the Eisenhower Matrix but treats it as a real product system: task state, scheduling, editing, storage, import/export, recurring work, and desktop ergonomics.
System Design
Quadrants are only the visible layer
The useful engineering work lives beneath the four boxes: persistence, task movement, reminders, schedule metadata, user preferences, and future classifier hooks.
- Separate urgent/important state from due and scheduled state.
- Support desktop-first speed without forcing cloud sync.
- Leave room for AI suggestions that remain explainable and reversible.
Engineering Challenge
Make automation feel controllable
For an AI direction to work here, the model should propose labels, priorities, and scheduling hints while the user stays in charge of final placement and commitments.
Next Iteration
Cross-platform rewrite with a cleaner product core
The next generation can move toward a TypeScript/Tauri architecture, keeping desktop quality while preserving a path to mobile and richer AI workflows.
Media