Personal AI Employee
Autonomous AI Agent with Claude Code & MCP
Autonomous AI agent system that manages real personal and business workflows. Monitors multiple communication channels (Gmail, WhatsApp, LinkedIn) using watcher scripts and converts incoming events into structured tasks automatically. Features a local-first vault workflow that prioritizes and routes tasks with human-in-the-loop approvals for sensitive actions. Supports MCP-based external action architecture enabling flexible integration with various tools and services. Provides comprehensive observability with live dashboard showing watcher health, queue state, and activity history. Implements autonomous execution loop with weekly CEO briefing reports. Silver tier complete with Gold-tier integrations in progress. Demonstrates deep hands-on understanding of AI automation, reliability, observability, and safe agent design patterns.
Why it mattered
Personal workflows live across email, WhatsApp, LinkedIn, notes, and task lists, which makes follow-up and prioritization fragile.
What shipped
A local-first personal AI employee that watches channels, converts events into tasks, routes them through an approval loop, and produces executive briefings.
Proof signal
Multi-channel event monitoring; Autonomous workflow management; Human-in-the-loop approvals; Live observability dashboard; Weekly executive briefings; Safe agent design patterns
System flow
Watchers collect events, a structuring layer turns them into tasks, a vault stores context, human approval gates risky actions, and MCP-style tools execute safe work.
Honest maturity snapshot
Manual questions, scattered context, and slow follow-up.
An agent workflow that routes intent, calls tools, and returns useful next actions.
Features
- Multi-channel event monitoring.
- Task structuring and prioritization.
- Local-first vault workflow.
- Human-in-the-loop approvals.
- MCP-style external action architecture.
- Weekly CEO briefing reports.
Engineering challenges
- Preventing automation from acting too broadly without approval.
- Maintaining observability across many small background watchers.
- Designing useful memory without creating privacy risk.
Results
- Shows advanced understanding of safe agent design.
- Positions the portfolio as a real AI systems lab.
- Creates reusable patterns for future client automations.
What I learned
- Tradeoff learned: Preventing automation from acting too broadly without approval.
- Proof learned: Shows advanced understanding of safe agent design.
- Next iteration: Add a public sanitized trace viewer.