Principal engineer · building since 1979
Legacy software that must keep running. Integrations where neither side tells the whole truth. Financial and AI-assisted systems whose results must be proved before they are allowed to matter.
The work in one sentence
The recurring problem is rarely a missing framework. It is an incomplete picture of the live system, an assumption nobody has challenged, or a failure that can still look like success. My job is to expose that boundary before it becomes the client’s loss.
Older desktop and low-code applications often contain years of operational truth that no specification captured. The work begins by finding the authoritative source, mapping the real rules, and planning a reversible transition—not by declaring a rewrite.
In a research system connected to broker infrastructure, a software check could not run and incorrectly reported success. Real execution remained disabled by a separate decision. That prevented the immediate loss—but it also exposed the danger of letting a temporary safeguard become permanent without an owner and a deadline.
A candidate cleared the configured floor but performed worse than the smaller model already doing the work. The business question was not whether the challenger was competent. It was whether the improvement justified hardware, power, integration, and operational cost. It did not.
Where I am most useful
Diagnose the live system end to end, identify the real failure boundary, and give the work one accountable owner.
Open the method →Deliver interface, services, schema, and third-party integration as one operating system—with a safe transition from what already works.
Open the casework →Use models to widen search and challenge designs while keeping architecture, data, acceptance, and release under engineering control.
Open the evidence →Two field reports
How an experienced engineer turns faster AI output into dependable software—and the six decisions that cannot be delegated.
Read the paper → HMAS field reportThe design, failed assumptions, audit, and reconstruction of an orchestrated multi-agent research system.
Read the HMAS paper →Experience in motion
Experience helps me recognize the pattern. Continuous learning keeps me from trusting it too quickly. Evidence and judgment determine what I do next.
I remain an active New Jersey Mobile Intensive Care Paramedic, currently serving a rural territory of more than 550 square miles. Patient care commonly lasts 30 to 90 minutes before handoff, so the work requires calm action, continuous reassessment, and a willingness to change the plan when the condition changes.
Maintaining that license requires continuing education as medical protocols change. I bring the same discipline to engineering: structured study, daily technical research, and working prototypes that turn promising ideas into evidence. MIT's 6.00.1x and AI TraceVector Alpha are two points on that same loop.
Field notes
A dual-architecture llama.cpp build worked until its folder was renamed. The executable had remembered an absolute library path—and turned a routine deployment into a fleet-wide failure.
Disaster-recovery teams prove their backups by rebuilding the whole company in a test facility. When a materially better tool arrives, your architecture deserves the same treatment: re-derive the system from scratch and see which decisions survive.
People ask how I dare run Chinese models on my own GPUs — while piping their entire codebase to a third-party API. The IP threat model is exactly backwards: a local file of weights can't exfiltrate anything. The wire can. And the wire has a second problem: getting Sherlocked.
Tell me what has to keep working, what is already promised, and what nobody has been able to explain. I will tell you whether I am the right person to own it.