DATA — Daily Autonomous Task Assistant

Managing responsibilities across three domains simultaneously — a corporate team of analysts and managers, five-plus leadership roles at church, and personal obligations as a father, husband, and caretaker for a parent — creates an overwhelming volume of communications, appointments, and tasks. Email inboxes were backlogged in the hundreds. Follow-ups were being missed. There was no single view of priorities across all domains, making it impossible to gauge actual throughput and capability within any given day.

The deeper problem wasn't just volume — it was the lack of visibility into what truly required attention versus what was noise. Traditional task management tools treated each domain in isolation, and email remained an unsorted stream of varying urgency.

This project began as a personal tool, which meant the discovery process was a conversation with myself — but it followed the same iterative pattern I use with stakeholders. The initial hypothesis was that consolidating tasks would be the key to executing across all responsibilities. As I began building, I discovered that communications were equally important. Then calendar obligations became the third pillar. The scope evolved iteratively: Tasks, then Communications, then Calendar, then an AI layer.

The critical insight was that a "single pane of glass" needed to work across domains. I needed to query by domain (just work, just church, just personal) or view everything together, and the system needed to understand the difference between a planned date, a target date, and a hard deadline.

Rather than adopting a commercial task manager and layering integrations on top, I built a purpose-designed system with three key innovations:

Three-Date Task Model: Every task carries a planned_date (when you'll work on it — auto-rolls forward when rescheduled), a target_date (original goal — never changes, tracks slippage), and a hard_deadline (external commitment — triggers alerts). A times_rescheduled counter auto-increments, and tasks exceeding a slippage threshold get highlighted. This was born from frustration with prior workflows that auto-rolled dates and hid slippage.

AI-Powered Email Triage: Instead of manually processing hundreds of emails, the system uses AI-scored attention ratings. Emails requiring user action are surfaced proactively. The user trains the system by marking emails as "needs attention," "done already," or "false flag." Over time, AI-generated rule suggestions based on existing rules automatically filter promotional and junk mail.

Integrated AI Planning: A planning skill identifies the crux of any task, breaks it into smaller pieces and deliverables. A research skill takes the plan and goes to the web for research. Complex tasks become manageable with a click of a button.

Tech Stack: React 18 + Vite (frontend), FastAPI Python 3.11+ (backend), Firestore (cloud DB), Smartsheet (task source), Gmail API, Google Calendar, Google Sheets, Claude AI (Anthropic), Playwright (E2E testing).

Deployment: Backend on Cloud Run (staging + prod), frontend on Vercel (staging + prod), Google OAuth authentication.

The application features 136+ API endpoints providing a single pane of view across all three life domains. App Script automations with configurable rules automatically clear email clutter — both historical and incoming. The system manages daily throughput and capability within a 24-hour period.

• All email inboxes brought to zero daily (from hundreds backlogged)
• Went from missing key emails and follow-ups to proactive management
• Auto-categorization of emails improves over time with user training
• Can manage actual throughput and capability within a 24-hour period
• AI planning reduced complex analysis tasks from weeks of team effort to hours of AI-assisted work
• Personal tool with security/privacy controls — demonstrates full-stack AI integration capability