I design engineering systems that convert abstract concepts into inspectable, reproducible, and operational artifacts.
My current work focuses on:
- low-level execution and memory-locality architectures
- AI runtime control planes with explicit authority boundaries
- desktop operation shells for agents, tools, permissions, approvals, audit, and recovery
- black-box systems treated as engineering targets rather than objects of blind trust
| Project | Role | Scope | Current Position |
|---|---|---|---|
| Project ORCHID | Concept Originator | Low-level micro-architectural execution core: cache-locality kernels, memory-role scheduling foundations, and execution-engine primitives | Public collaborative development; implementation and architectural expansion led by Kevin West |
| GUI Shell | Architect / Maintainer | Generic desktop-first AI Runtime / Agent Operation Shell with adapter, permission, approval, audit, and recovery boundaries | Owner-use operation complete; OSS/product release claims remain gated |
| BLUE-TANUKI | Architect / Maintainer | Local resident AI control plane with deterministic authority handling and downstream LLM/tool execution | v1.0 Release Candidate; public GA remains subject to explicit release validation |
Project ORCHID began from my initial concept implementation for CPU cache-line locality, assembly-generation matrices, and parallel multi-memory-bank role scheduling.
The project is now being developed publicly within the DigitalServerHost organization, with:
- Concept Originator: Teppei Oohira / 大平鉄兵 (@gatchimuchio)
- Core Architecture & Maintainer: Kevin West (@westkevin12)
ORCHID is currently progressing from a low-level proof-of-concept toward a broader execution-engine architecture. Its public repository preserves the original conceptual foundation while separating ongoing implementation and maintenance responsibility.
Repository: DigitalServerHost/ORCHID
GUI Shell is a generic desktop-first operation shell for local AI runtimes, agents, tools, and services.
Its design principle is simple:
The UI may display and request actions, but it must not create authority.
Core concerns include:
- adapter conformance
- authority stripping
- content exposure boundaries
- permission and approval control
- audit evidence
- recovery surfaces
- bounded native operations
Repository: gatchimuchio/GUI-Shell
BLUE-TANUKI is a local resident AI control plane designed around explicit authority ownership and inspectable execution boundaries.
Its governing principle is:
LLMs, tools, schedules, and channels are downstream devices. They do not own authority.
Current public work includes:
- resident Control Center
- runtime snapshots and authority traces
- approval gates
- hash-chain audit logs
- bounded tool execution
- channel adapters
- guided first-run operation paths
Repository: gatchimuchio/blue-tanuki
I am primarily interested in systems that remain valid after contact with reality.
My working principles are:
- Structure over noise
- Reproducibility over impression
- Explicit boundaries over ambiguity
- Safety before expansion
- Operational evidence before completion claims
- Implementation responsibility must be separated from conceptual origin where appropriate
The repositories above are public engineering artifacts under active development.
- ORCHID is a public collaborative execution-engine project derived from my initial conceptual implementation.
- GUI Shell is operational for owner-use development stages, but does not claim completed product release readiness.
- BLUE-TANUKI is at Release Candidate stage, but does not claim public GA without explicit validation and approval.
The objective is not to present unfinished work as finished products.
The objective is to build public, inspectable engineering systems whose progress can be verified through code, documentation, tests, and operational boundaries.
- GitHub Profile: gatchimuchio
- Project ORCHID: DigitalServerHost/ORCHID
- GUI Shell: gatchimuchio/GUI-Shell
- BLUE-TANUKI: gatchimuchio/blue-tanuki