Notes from the engineering bench

Make the work easier to reason about.

Short field notes on qualification, engineering evidence, and focused AI and data transformation. Each note starts from a decision an engineering team needs to make.

Engineering notes

A prototype proves a possibility. A system makes it repeatable.

A successful prototype answers an important question: can the idea work? Product development then asks harder questions. Can another engineer reproduce the result? Can the design be reviewed, qualified, maintained, and changed without losing the evidence behind earlier decisions?

The practical shift is to treat architecture, simulation, laboratory work, and verification as one evidence chain. Start with the decision that must survive, identify the evidence it needs, then make ownership and review points explicit. The goal is not more process; it is fewer heroic recoveries.

See the evidence in selected work Discuss a scaling constraint

Simulation becomes infrastructure when its assumptions are reviewable.

A simulation file can produce a convincing plot and still leave the team dependent on the person who created it. Inputs, assumptions, versions, and expected outputs need to travel with the model if the result is going to support a product decision.

Small automation layers can turn expert-only runs into repeatable studies. The useful measure is not how much software surrounds the model; it is whether an engineer can rerun the study, inspect the assumptions, and explain why the result is trustworthy. The pyPLECS and pyFEMM records show this pattern in practice.

Explore the portfolio projects How I handle evidence and confidentiality

Give AI a bounded engineering decision, not a vague mandate.

AI is most useful in engineering operations when the target decision, available evidence, and acceptable failure modes are clear. That might mean comparing design alternatives, finding gaps in a review package, or reducing the manual work around a simulation workflow.

Begin with the bottleneck and a baseline. Keep an engineer responsible for the decision, expose the source evidence, and measure whether the workflow becomes more reliable or merely more impressive. A focused tool should earn its place inside the engineering system.

Read the AI and data FAQ Write about a concrete bottleneck

The continuing archive

Follow new notes on Substack.

Substack is the only newsletter subscription destination. This site does not ask for or store your email address.

Read and subscribe on Substack