Autonomous Systems
Multi-agent coordination and real-time adaptive control — secure systems where specialized agents work together within strict isolation boundaries.
Contents
Principles
- Security by Design: Every agent runs in its own container with minimal privileges. No direct agent-to-agent communication — all coordination flows through the NATS broker.
- Clear Coordination: The orchestrator uses cloud LLM reasoning for task planning while agents stay lightweight and deterministic.
- User Control: Sensitive actions require explicit approval. The system explains its reasoning and provides a clear audit trail.
- Composable Architecture: New capabilities are added by deploying new agent containers, not by modifying the orchestrator.
How They Work Together
LLM reasoning · task planning"] ORCH <--> NATS["NATS Broker"] NATS <--> HA["Health Agent
biometrics"] NATS <--> HOME["Home Agent
HomeKit"] NATS <--> MON["Monitoring Agent
backups · infrastructure"]
Each agent subscribes to the NATS topics matching its domain. The orchestrator breaks a complex request into domain-specific tasks, publishes them, and aggregates the results. Agents can trigger other agents through the broker but never communicate directly.
Systems
Training Assistant
Recently Updated · Swift · Full Details →
Native SwiftUI training application for iPad and Apple TV that bridges a Wahoo KICKR Core smart trainer with AI-driven workout logic. The app communicates with the trainer over Bluetooth (FTMS protocol) for real-time resistance control and telemetry, while a NATS message bridge connects to an external agent for dynamic workout decisions.
Capabilities:
- Real-time power, cadence, and heart rate telemetry
- Dynamic resistance adjustment based on training goals
- Adaptive interval programming via agent reasoning
- Session recording and performance tracking
| ← Back to Projects | Signal Processing | Photography | Data & Analytics |