2025 Featured

SAVANT Intelligence System

A comprehensive personal AI-powered life operating system integrating health, productivity, and cosmic timing across Mac, iOS, and Raspberry Pi.

Technologies

TypeScript Node.js Claude API BigQuery SwiftUI Raspberry Pi MCP Protocol

Key Outcomes

  • 7 custom MCP servers handling different life domains
  • Real-time sync across Mac, iOS, and Pi
  • AI-powered check-ins that adapt to energy levels
  • Integrated health tracking with Apple HealthKit

Building a Personal Intelligence Layer

SAVANT (Savant Autonomous Visionary Ambient Navigation Tracker) is an AI-powered personal operating system that I built to integrate every aspect of daily life into a coherent, actionable intelligence layer.

The Architecture

The system operates across three devices in a mesh network:

Mac (Primary)

  • Claude-powered conversational interface
  • 7 custom MCP servers for different domains
  • BigQuery data warehouse for historical analysis
  • Obsidian vault integration for knowledge management

iOS (Mobile)

  • SwiftUI app with conversational check-ins
  • HealthKit integration for automatic health data
  • Offline-capable with sync on reconnect

Raspberry Pi (Edge)

  • Discord bot for remote access
  • Partner dashboard (shared context)
  • State mirroring for redundancy
  • Tailscale mesh networking

Key Features

Adaptive Check-ins The system adjusts its questions based on current energy levels, cycle phase, and cosmic timing. Low energy days get shorter, gentler check-ins.

Multi-AI Orchestration Uses a Quad-AI architecture: Claude for building, Gemini for pattern analysis, Codex for scaffolding, and a local model for private data.

Evidence Harvesting Automatically surfaces wins and achievements from daily notes, formatting them as STAR stories for portfolio building.

Technical Highlights

  • Custom MCP protocol implementation
  • Real-time WebSocket sync across devices
  • BigQuery integration for long-term pattern analysis
  • Automated coherence scoring across all data layers

This project represents thousands of hours of iteration on the question: “What would a truly intelligent personal assistant actually look like?”