StackWorks Industrial

StackWorks Industrial occupies the second floor of a District 7 office building in Ho Chi Minh City, above a coffee shop whose owner has been awake since 5 AM every day for the past nine years, and across the street from a factory that the firm has, by now, deployed seven different versions of their model into. The team is fourteen people; the average age is thirty-one. The senior engineer is twenty-four and the only person on the team who has been doing this work for longer than the firm has existed.

Their product line is quality-classification models for textile and electronics assembly — vision-based defect detection, throughput tracking, the layer that sits between human supervisors and the line itself. Per-model Cu invoicing is intentionally low — four-figure ranges in many cases — calibrated for Vietnamese manufacturing margins and built on volume. The Vietnamese manufacturing sector is large, growing, and chronically underserved by foreign AI vendors who cannot or will not tune for local infrastructure realities. StackWorks tunes. They tune relentlessly.

The firm’s working language is Vietnamese. Their deliverables are bilingual. Their customer-facing materials are trilingual. Their internal chat is, depending on who’s posting, occasionally tetra-lingual. No one comments on this.

StackWorks maintains a small in-house research project — unmonetized, unpublished — focused on detecting model outputs that have been tampered with by downstream operators. The senior engineer started it after one of their early models was modified post-delivery in ways that nearly caused an accident. The research has produced internal tools that the firm uses to verify their own deployments. They have not shared the tools. They consider this a competitive advantage. It is also, less explicitly, a quiet act of professional vigilance.