Approach
UX leads. AI builds.
Users validate.
Every project I take on follows the same core principle: understand the people first, then decide what to build. Technology choices come last — after I have observed the actual workflow, identified the real pain points, and validated that a solution will genuinely help.
This approach has taught me that the most valuable AI systems are often invisible to their users. They remove friction, handle tedious steps, and let people focus on the judgments that require human expertise.
Much of my work starts in unfamiliar territory — a new industry, an undocumented workflow, a team under pressure to move faster. I have learned that the hard part is rarely the model choice. It is getting from messy reality to a system people actually trust and use. The four phases below are how I do that, consistently.
Listen & Observe
I start by interviewing the people who do the work — process owners, operations staff, domain experts. I watch how they actually complete their tasks, noting workarounds and pain points that are invisible in process documentation. At Ergomotion, this revealed that customs brokers had developed manual shortcuts that no system had captured.
Prototype & Test
Rather than building a complete system, I ship a working prototype as fast as possible and put it in front of real users. Their reactions tell me more than any specification document. For the Pallet Calculator, I built the same capability two different ways — deterministic and AI-driven — to learn which approach users actually trusted.
Build with the Right Tool
AI is powerful, but it is not always the right answer. I choose the cheapest, most reliable tool for each step — deterministic logic for rules-heavy calculations, small language models for simple parsing, large models only when the task demands it. For the Pallet Calculator, this meant deterministic geometry for the layout math and a small model only at the edges — faster, cheaper, and easier to debug than an LLM-everywhere approach.
Close the Loop
I train the people who use the system to be its testers. Their ongoing feedback becomes the quality assurance layer. At Ergomotion, process owners became the QA team, and their feedback shaped the final automation pipeline. The system improves because the people closest to the work are continuously involved.
The constant is not the tool — it is getting from messy reality to a system people trust and use.