About
Toland Advisory is my practice. When you engage it, you get me, not a team you never meet and a junior consultant on site.
Joseph Toland
Principal enterprise architect, 30+ years across enterprise, application and data architecture. I set reference architectures, chair architecture review boards, and run ADR-based governance in regulated environments, while staying close enough to delivery that the direction survives contact with implementation.
Recent work has been about bringing AI platforms under the same discipline as everything else: workload identity for agents, governed retrieval, guardrails, evaluation, and cost you can attribute to a feature.
How I work
Evidence over assertion
Findings are tied to artifacts I asked to see, not to what an interview claimed. The gap between the two is usually the finding.
Still a working engineer
.NET and C#, Python, and Linux. The architect who earns trust is the one willing to open the terminal and validate the design rather than hand it off.
Bounded and fixed price
Small first engagements with published scope. You should be able to stop after one and still have something worth the money.
Building, not just advising
I build agentic systems in production, not slideware about them. LangChain, LlamaIndex, CrewAI, AutoGen, LangGraph and MCP, with RAG and vector pipelines against the Anthropic and OpenAI APIs, across AWS, Azure and GCP.
I am also the solo founder of an AI health platform, which means I carry the full loop myself: shipping it, paying for the tokens, watching the costs, and feeding real usage back into the roadmap. That experience is why cost control and evaluation show up in every architecture I design rather than getting deferred to a later phase.
The current build is Cloud Well-Architected Framework, a landing zone factory that generates Azure infrastructure against the Well-Architected Framework. What it does today is compose an architecture request into Terraform, with the cost of each decision projected before you commit to it. What it is being built toward is enforcing conformance at the moment the artifact is generated instead of scoring it afterwards, which is the difference between drift being rare and drift being routine. That enforcement is designed and not yet written. It is the argument I make to clients, built as code I have to make work rather than a slide I have to defend. The build note is more specific.
Michigan, remote
Based in New Baltimore, Michigan, and working remotely across the US. Onsite in southeast Michigan on request, and travel elsewhere by arrangement.