Personal AI Employee — Hackathon 0

Personal AI Employee is a controlled, auditable AI operations system built with Human-in-the-Loop (HITL) governance, workflow state management, MCP integrations, and structured audit logging.

This project is not a chatbot.
It is a state-driven AI workflow system designed to convert unstructured inputs into controlled, reviewable, and traceable execution workflows.

---

Live / Demo Links

• UI Demo: https://personal-ai-employee-ui.vercel.app/dashboard.html
• Final Proof Branch: platinum-final-demo

Note: The main branch is used as the public project overview. The final demo-ready implementation and evidence are available in the platinum-final-demo branch.

---

Final Review Branch

The final working demo and proof branch is:

platinum-final-demo

The main branch is used as the recruiter-facing project overview.
The full final implementation, evidence flow, UI, and demo-ready system are available in the platinum-final-demo branch.

---

What This Project Demonstrates

Personal AI Employee demonstrates how AI automation can be made safer through:

• Human approval before sensitive actions
• Controlled task lifecycle management
• Deterministic workflow states
• MCP-backed tool integrations
• Append-only audit logging
• Evidence-based execution proof
• CEO briefing/report generation
• Runtime visibility and state tracking

The system is designed around safety, traceability, and accountability.

---

Core System Flow

Input → Task → Plan → Human Approval → Execution → Logs → Reporting

Full demo flow:

MCP → Runtime → Watcher → Task → Plan → Approval → Artifact → Logs

Each step produces verifiable files, artifacts, or logs inside the project vault.

---

Architecture Overview

1. Input / Perception Layer

The system accepts tasks through watcher-based intake.

Implemented intake sources include:

• Filesystem watcher
• Gmail intake bridge
• Manual task creation

Raw inputs are converted into structured tasks.

---

2. Vault-Based State Machine

The project uses a local vault as the system memory and workflow engine.

Main workflow states:

Needs_Action → In_Progress → Plans → Pending_Approval → Approved → Done

The vault acts as:

• Source of truth
• Workflow state tracker
• Audit-friendly system memory
• Human review surface

---

3. Planning Layer

The system generates structured execution plans before any action is taken.

Plans are reviewable and do not automatically execute sensitive actions.

---

4. Human-in-the-Loop Governance

Human approval is required before controlled execution.

Key rule:

No approval → no execution

This ensures the system remains safe, reviewable, and accountable.

---

5. Controlled Execution Layer

Approved actions can be executed through controlled integrations.

Implemented / demonstrated integrations include:

• Slack
• Odoo
• Email
• Filesystem
• Gmail
• GitHub
• Google Calendar
• Playwright
• Context7

All sensitive actions are governed by approval and logging.

---

6. Audit and Evidence Layer

The system keeps structured logs and evidence for verification.

Evidence includes:

• JSONL audit logs
• Watcher logs
• Execution logs
• Approval artifacts
• CEO briefings
• Demo screenshots
• Platinum evidence pack

This makes the workflow traceable from input to final output.

---

Tier Progression / Branch Guide

Branch	Purpose
main	Recruiter-facing overview README
silver-dev	Silver watcher/perception layer
gold-dev	Gold orchestration, audit, and CEO briefing layer
gold-gap-fix	Gold repair and finalization branch
platinum-start	Start of Platinum visibility layer
pipeline-repair-safe	Safe pipeline repair branch
platinum-final-demo	Final working demo and proof branch

---

Gold Layer

The Gold layer proves the core execution system.

It demonstrates:

• Deterministic workflow execution
• Resume-safe processing
• Human approval gates
• Structured audit logs
• CEO briefing generation
• Controlled execution after approval
• Retry and crash-safety concepts

Gold represents the main controlled execution system.

---

Platinum Layer

The Platinum layer adds visibility and final demo readiness.

It demonstrates:

• Runtime visibility dashboard
• Cloud vs local execution separation
• Clear state ownership
• Final UI/demo presentation
• Evidence pack organization
• Final proof branch preparation

Platinum improves observability and evaluation clarity without weakening the HITL safety model.

---

Key Safety Guarantees

• No uncontrolled execution
• No sensitive action without approval
• Human-in-the-loop enforcement
• Append-only audit logging
• Traceable workflow transitions
• Deterministic state movement
• Clear Gold vs Platinum boundaries

---

Evidence Locations

Important evidence and proof files are organized across the final branch, especially:

AI_Employee_Vault/
AI_Employee_Vault/Logs/
AI_Employee_Vault/Artifacts/
AI_Employee_Vault/Briefings/
AI_Employee_Vault/Artifacts/PLATINUM_EVIDENCE_PACK/
docs/
evidence/

The final demo-ready evidence is available in:

platinum-final-demo

---

Quick Demo Concept

A judge-safe demo follows this flow:

1. Drop or receive an input task
2. Watcher converts input into a structured task
3. System generates a plan
4. Plan moves to pending approval
5. Human approves the action
6. Controlled execution runs
7. Output is saved as an artifact
8. Logs record the full lifecycle

---

What Makes This Project Different

This project is not a simple automation script.

It is a controlled AI operations system focused on:

• Safety
• Auditability
• Human approval
• Real workflow states
• Tool integration
• Evidence-backed execution

The goal is to show how AI automation can be made practical, governed, and reviewable.

---

Current Status

Final project status:

Gold + Platinum demo system completed
Final proof branch: platinum-final-demo
Live UI demo available

This project demonstrates practical AI automation engineering, workflow orchestration, HITL governance, MCP-style integrations, and audit-ready system design.