/projects

Things I've built.

Not clones. Not tutorials. Real systems that I use every day — designed, architected, and deployed by me on hardware that fits in my pocket.

🦞01

The OpenClaw Agent Ecosystem

Multi-agent AI system on a single Raspberry Pi

The Problem

Most AI demos are single-purpose chatbots. I wanted a system where multiple specialized AI agents work together like a team — reviewing code, managing tasks, controlling a home, and handling daily workflows.

The Solution

Built a multi-agent orchestration system using the OpenClaw framework on a Raspberry Pi 5. Multiple specialized AI agents communicate, delegate, and execute tasks autonomously — powered by cloud AI and local models running side by side.

Key Results

  • Runs 24/7 on an $80 Raspberry Pi 5 with 8GB RAM
  • Multiple specialized agents with distinct roles: orchestrator, reviewer, task manager, data analyst
  • Cloud AI models for complex tasks, local models for private data
  • Full code review pipeline with automated peer review
  • Integration with Home Assistant, Fitbit, Google Drive, and Twilio
Raspberry PiMulti-AgentDockerAgent ArchitecturePython
🤖02

The Orchestrator Agent

The AI that runs everything

The Problem

Managing multiple AI agents, APIs, smart home devices, calendars, and notifications from a single point requires an orchestrator that can think, delegate, and execute.

The Solution

The primary orchestrator agent manages the entire system. It coordinates the agent team, controls home automation, handles email and calendar, monitors health data, and communicates across Telegram, web, and voice channels.

Key Results

  • Handles 20+ integrated services simultaneously
  • Auto-routes tasks to specialized sub-agents based on skill and workload
  • Voice calls via Twilio with custom personas
  • Background heartbeat system for proactive monitoring
  • Full memory system with daily logs and long-term recall
Multi-AgentOrchestrationTwilioAPI IntegrationNode.js
🧛‍♂️03

The Code Review Agent

An energy vampire that catches your bugs

The Problem

Code reviews are easy to skip when building solo. But skipping them means bugs ship, architecture suffers, and technical debt piles up.

The Solution

A dedicated AI code reviewer that examines plans with excruciating detail before any code gets written. It evaluates edge cases, risks, and architectural concerns — producing structured verdicts that catch problems early.

Key Results

  • Reviews plans before execution — catches issues early
  • Tediously thorough edge-case analysis
  • Produces structured verdicts: Summary → Risks → Recommendations
  • Also serves as backup orchestrator if primary goes down
  • Direct Telegram channel for on-demand access
Code ReviewAI AgentPrompt EngineeringAutomation
🏛️04

The Task Board

Kanban for AI agents

The Problem

Standard kanban boards don't route tasks to AI agents. I needed a task system where agents could be assigned work, collaborate, and report progress — all autonomously.

The Solution

Built a full kanban board that auto-routes tasks by type and skill. Tasks flow through stages — from intake through planning, collaboration, execution, and review — with automatic routing to the right agent.

Key Results

  • Auto-routes code reviews to reviewer, data tasks to analyst
  • REST API for agent integration
  • Clean kanban UI with drag-and-drop
  • Tracks task history and agent performance
FastAPIReactSQLiteTask RoutingAPI Design
🏠05

Smart Home, Supercharged by AI

Voice-controlled house with an AI brain

The Problem

Smart home hubs are siloed and rigid. I wanted natural language control across lights, appliances, climate, garage, irrigation, and entertainment — all accessible through AI.

The Solution

Integrated Home Assistant with the agent system. Now the orchestrator can control every smart device through natural language. Lights, thermostat, garage door, washer, dishwasher, speakers, TVs — all accessible via chat or voice.

Key Results

  • 40+ smart devices across multiple rooms
  • Natural language control via Telegram
  • Automated routines based on time, presence, and weather
  • Integrated with wearable health data for health-aware automations
Home AssistantIoTTelegram BotAutomationREST API
06

Infrastructure on a Credit Card

The entire stack on a Raspberry Pi 5

The Problem

Everyone thinks you need a data center to run AI. I wanted to prove you can run a serious production system on hardware that costs less than dinner out.

The Solution

The entire infrastructure runs on a single Raspberry Pi 5 with 8GB RAM and a 2TB SSD. Docker containers for services, reverse proxy with automatic SSL, mesh VPN for secure remote access, and everything tied together by the gateway.

Key Results

  • Everything on one Raspberry Pi 5 — agents, services, web server, database
  • Docker containers for isolation and easy management
  • Nginx reverse proxy with automatic SSL certificates
  • Mesh VPN for zero-config remote access
  • Nightly encrypted backups to cloud storage
Raspberry Pi 5DockerNginxTailscaleLinuxDevOps