Building applied AI systems for healthcare and agent-powered work.
I'm a founding engineer at Architect Health working on AI and data infrastructure for healthcare. I care about turning AI from impressive demos into tools people can trust in day-to-day work.
My work sits where applied AI, healthcare operations, agent systems, and developer infrastructure meet. This site is a working notebook on building useful systems carefully, with attention to reliability, privacy, and the people affected by the workflow.
Selected work
Agent products
A product model for agents that can plan, build, review, resume, and keep a person clearly in the loop.
What it taught me: Useful AI work needs structure around the model: ownership, review, and a clear record of what happened.
Read moreReliability
A framework for turning agent behavior into something inspectable, repeatable, and safe enough for production workflows.
What it taught me: Reliability is not a prompt. It is a product surface, an architecture, and an operating model.
Read moreInfrastructure
Local-first and hybrid inference patterns for products where cost, privacy, uptime, and model control affect customer trust.
What it taught me: When AI is central to the product, infrastructure choices become part of the customer promise.
Read moreDecision systems
Tools that give agents current, structured information so they can support decisions with clearer evidence.
What it taught me: A product is only as useful as the quality of the information it can act on.
Read moreEvidence
Screenshot feedback loops that help agents verify user-facing work before asking a person to approve it.
What it taught me: For AI work to be trusted, the system needs evidence, not just completion messages.
Read moreWhat I am learning
Some come from agent systems. Some come from healthcare and infrastructure. The pattern is that real products bring the business, operational, and technical questions together around the people doing the work.
See longer notesWhen AI touches real workflows, quality, auditability, and recovery paths matter as much as the model itself.
The hard part is deciding what the system should know, when it should know it, and who is accountable for the result.
The distance between a promising demo and a trusted product is mostly operations: permissions, reviews, retries, and evidence.
AI can help small teams take on more work, but only when the surrounding system keeps the work accountable.
Latest notes
When AI is central to the product, model access, data boundaries, routing, cost, and reliability become important infrastructure decisions.
Prompts help, but reliable AI products need operating rules, review loops, permissions, and failure handling.
As agents take on more execution, the work shifts toward clearer goals, careful review, and accountable decisions.