Working notes, methods, and lessons from production engagements — written by the engineers who shipped them. Practical, opinionated, occasionally contrarian.
A representative cross-section — half engineering practice, half methodology, with the occasional contrarian read on industry conventional wisdom.
Most predictive-maintenance pilots produce decent models and almost no operational impact. The reasons are rarely technical — and the patterns repeat across industries.
→Trust is engineered, not requested. A breakdown of the design patterns that move conversational and agentic AI systems from "interesting demo" to systems frontline workers rely on.
→Multi-echelon inventory optimization is the longest-promised, least-deployed capability in modern supply chains. A working theory of why — and what changes when AI is paired correctly with OR.
→RL controllers vs PID — when does it actually pay? An honest accounting from coating, extrusion, and curing-process deployments where the answer was sometimes "yes" and sometimes "stop reading hype".
→How to move sensor and MES data into cloud lakehouses without compromising plant security or breaking Purdue Model boundaries. A reference pattern, with notes from real deployments.
→The last seven percentage points of inspection accuracy are the hardest, and they are where ROI lives. A walkthrough of the engineering work between adequate and excellent CV inspection.
→Our most-requested engagement, deconstructed. The five-stage pipeline that turns decades of paper into a working multilingual conversational assistant.
→Six Sigma works. AI accelerates it dramatically. A practitioner’s view of where to inject AI in each DMAIC phase and where to leave the methodology alone.
→One issue per month. Working notes from current engagements. No newsletter platitudes.
Request subscription →