Joel Johnston
AI Operations Architect
I build the system AI runs inside.
What I Do
I design and build AI operations infrastructure — the orchestration layer between AI models and the systems that use them. Multi-model routing, capability-based dispatch, fault tolerance, quorum consensus, workflow orchestration. Not just calling APIs — architecting the platform that manages AI as a distributed workforce.
My current project, robonet, is an AI agent orchestration system that provides unified provider abstraction across 8+ AI platforms, capability-based task routing across a mesh of worker nodes, and quorum consensus for high-stakes AI validation.
Three Layers, Not One
Operator
Multi-model user — Claude, GPT, Grok, Gemini. Cross-validates outputs across models. Tracks error rates per model. Treats AI as peer, not oracle.
Infrastructure Builder
Builds the system AI runs inside. Provider abstraction, capability dispatch, quorum consensus, workflow orchestration. AI agents as distributed workers.
Internals Practitioner
Built an artificial neural network from scratch (~2011, pre-framework). Forward prop, backprop, gradient math, weight initialization, convergence. The internals, not just the API.
AI Operations Capabilities
| Capability | Implementation |
|---|---|
| Multi-model orchestration | Unified interface across Ollama, Groq, HuggingFace, Cloudflare, Anthropic, OpenAI, Google, xAI |
| Model routing & fallback | Capability-based dispatch — route by what workers can do. No single machine needs all API keys. |
| AI validation | Quorum consensus — fan out to N LLMs independently, synthesize consensus. No single model trusted for high-stakes decisions. |
| Agent lifecycle | YAML-based task graphs with dependency resolution, parallel execution, timeout enforcement, graceful degradation |
| Cost & rate management | Free tier vs premium routing. Rate limit distribution across mesh. Fault tolerance from topology. |
| Security | TLS PKI, challenge-response auth, engagement framework with rule-gating |
Technical Domains
10 technical domains at practitioner depth or deeper. Not 10 separate specializations — one cognitive operation (pattern decomposition) applied across domains.
- Distributed Systems Architecture
- Network Protocols & Transport
- Cryptography & Security
- AI / Machine Learning
- Enterprise Infrastructure & Cloud
- DevOps & Build Systems
- IoT / Embedded Systems
- ERP / Business Systems
- Web Application Development
- Systems Architecture (meta-domain)
Interests & Adjacent Domains
Domains outside IT where the same pattern-decomposition applies. Not professional claims — real depth from sustained engagement.
- Theology (pastor-level depth, covenantal focus)
- Multi-Dimensional Theoretical Physics
- Linguistics (10 human languages)
- Psychology / Cognitive Architecture
- Medicine (diagnostic pattern recognition)
Enterprise Background
Wells Fargo — PMG
Performance Management Group — 40 engineers selected from 1,000+ IT staff. C-level escalation oversight for failing projects. SOX/PCI regulated, multi-datacenter, zero-downtime environment.
Diagnosed a 2-3 year unresolved performance problem in approximately 3 days — with no prior Apache experience. Tens of millions in losses stopped.
US Bank — Application Consultant 4
Official title: C# developer. Actual role: the person people bring unsorted problems to. Replaced vendor product security with bank security on 2 projects. Trained senior staff across specialties.
Same "go-to" pattern emerged independently at two separate Fortune 500 financial institutions.
Technical Stack
Primary
- Python
- C#
- Bash / Shell
- JavaScript / TypeScript
Infrastructure
- Kubernetes
- Docker
- KVM / libvirt / Hyper-V
- Jenkins / CI/CD
AI Platforms
- Anthropic (Claude)
- OpenAI
- Google (Gemini)
- xAI (Grok)
- Ollama / Groq / HuggingFace
Background
Started coding BASIC on a Commodore 64 in the early 1980s. Ran a BBS and served as FidoNet regional coordinator in high school. Campus network admin and CS lab assistant at UW-La Crosse (Sendmail MTA, NeXTSTEP maintenance). IBM test school for Java (pre-1.0) and Eiffel. Adjunct instructor at a Wisconsin public technical college.
Currently building robonet and the xsubi platform independently — AI agent orchestration, virtualization infrastructure, build systems, and ERP modules across 6 interconnected projects.