Twelve-plus years of hands-on quality engineering — automotive, mechatronics, electronics fabrication, heavy equipment — now paired with data science, local LLMs, and the infrastructure engineering to run them on the shop floor. Proven systems first. New technology where it earns its place.
The core discipline: supplier & vendor quality, NPI launches, root-cause investigation, QMS design and audits, continuous improvement. Twelve years across regulated, high-stakes manufacturing.
Practical AI where it pays for itself: RAG systems over your technical documentation, local LLMs for audit prep and reporting, faster data analysis. Runs on-prem — your process data never leaves the building.
Getting quality data out of binders and into systems: digitalized inspection records, dashboards, vision-inspection pilots, additive manufacturing feasibility. Scoped pilots before big bets.
Senior Vendor Quality Engineering Specialist with 12+ years in the field — currently leading vendor quality at a machine-vision manufacturer, previously quality engineering roles across automotive, robotics, and electronics fabrication.
Alongside the day job: an R&D lead on metal additive manufacturing feasibility, mixed-reality training development, and building RAG systems and local-LLM tooling for real quality workflows — backed by a DevOps/infrastructure background (ASQ-CQE, UofT DSI AI/ML). The through-line is the same — bridging proven quality systems and the technology that comes next.
Proofs of concept and tooling built to test where AI genuinely helps quality work. Honest about maturity: proofs of concept are labeled as such; the tools run fully local — no cloud required. Full list at github.com/ALEX8642.
ResNet-18 on WM-811K wafer maps — focal loss, CBAM attention, Grad-CAM++ explainability. Macro-F1 0.916 across 9 defect classes. Extended with SimCLR self-supervised pretraining on 638k unlabeled maps (wafer-ssl).
Turns dual-microphone meeting recordings into clean, deduplicated transcripts and LLM-generated audit reports. Fully local — no data leaves the building.
Retrieval-augmented chatbot over technical documentation, built on Haystack 2.x and Ollama. Ask the manual instead of digging through it — runs entirely on local models.
Thirty minutes, no pitch. We look at your process, and I tell you honestly what's worth automating — and what isn't.