Computer-Vision Test Automation

Problem

Game and localization testing relied on people repeatedly inspecting on-screen visual states by hand. That work was slow, repetitive, and hard to run consistently at scale, which made it a strong candidate for automation.

Constraints

The automation had to reproduce judgments a human tester would make about on-screen state, work across game and localization testing workflows, and fit into QA processes that several teams already depended on. It also had to be repeatable enough to trust in regression runs rather than one-off checks.

Architecture

I created Python automation tools for repeatable game and localization testing workflows, using OpenCV to detect and evaluate on-screen visual states that previously required manual inspection. Working with QA, localization, engineering, and project-management teams, I identified the highest-value opportunities to automate and led the technical direction for those cross-functional projects.

Tradeoffs

Validating behavior through computer vision trades a degree of robustness for reach. Visual detection can be sensitive to changes in rendering or layout, so the automation needs maintenance as the games evolve. In return, it can check states that would otherwise require a person watching the screen. I focused the automation on the high-repetition, high-value checks where that trade clearly paid off, and left genuinely judgment-dependent scenarios to human testers.

Testing

The value of the automation depended on it being repeatable and trustworthy, so I validated the vision-based checks against the states a human tester would confirm by hand before relying on them in regression runs. Increasing the repeatability of regression testing was itself part of the goal.

Results

• Reduced manual QA effort and operating costs by approximately $161k per quarter docs/profile/projects.md.
• Increased the repeatability of regression testing docs/profile/projects.md.
• Allowed QA teams to focus more time on scenarios requiring human judgment docs/profile/projects.md.

What I'd Change and What I Learned

Vision-based automation earns its keep on repetitive, high-volume checks, but it carries a maintenance cost as interfaces change. What I would carry forward is to invest early in making the visual checks resilient and easy to update, and to keep choosing automation targets by value rather than by what is merely automatable. Leading these projects across QA, localization, and engineering, and over time managing teams of up to 40 people, also taught me that technical direction only lands when it is chosen together with the people who run the tests every day.