Applied Systems Architecture

Systems & Infrastructure

Autonomous development setups, edge routing engines, and predictable prompt compilers built to run in production.

Active FrameworkMulti-AgentGit Worktreesv20.0

The Moonglade Method

A cognitive operating system for pair programming with frontier AI models. It coordinates specialized tasks across isolated git branches, freezes system architecture before any code gets touched, and demands working terminal test proof before code merges into the main project.

Dual-Gearbox Modes

Shifts cleanly between clinical systems engineering and creative brainstorming without corporate polite padding.

Steered Latent Reasoning

Guides the model's native thinking budget toward finding race conditions, schema breaks, and edge cases before writing code.

Hard Test Gates

Never accepts assumptions. Demands real automated test execution and clean terminal runs before marking work complete.

Read Method Philosophy →

Prompt CompilationDiffusion Models1-Shot Invariants

Persona-Grounded Visual Anchoring (PGVA)

A four-tier prompt compilation architecture that solves character drift and prompt fatigue in generative image tools. By separating physical facial anatomy and boundary rules from simple scene intent, you can generate consistent character portraits across realistic photography, Studio Ghibli watercolors, and retro sci-fi using short, three-word prompts.

Tier 1: Invariant Core

Static bone structure, skin tone, and ethnic demographics.

Tier 2: Signature Sigils

Distinctive jewelry and lighting termination rules.

Tier 3: Wardrobe Sets

Environment-matched clothing archetypes.

Tier 4: Scene Intent

Minimal action, time of day, and lighting notes from the user.

Read Engineering Case Study →Download Builder's Guide (.md) ↓View Generated Gallery →

Edge InfrastructureCloudflare WorkersTypeScript

Resilience at the Edge: Discord AI Router

A reliable, fast AI message routing engine built for Discord. It pairs stateless Cloudflare Workers at the edge with a stateful Node.js runner to provide automatic failover across model providers, offline message handling, and sub-second reply times.

Fast Global Starts

Global distribution via V8 edge isolates for immediate message dispatching.

Decoupled State

Clean state handling that recovers gracefully from local power or network drops.

Automatic Failover

Dynamic fallback routing across Anthropic, Google, and OpenAI backends.

Read Architecture Breakdown →