Ray Browser was a Chromium-based browser with split-screen features, a Ray App Studio for creating apps, and a Ray Agent to talk with (an agentic wrapper) that accumulated a memory for your Ray Profile.
I spent the last six months heavily pivoting to research & development — studies on how the browser could be made to work with you, instead of being the age-old "browser that never changes." That kicked out a lot of prototypes and ideas, and I'm still on this road of thinking: how could automation and workflow really be used to bring automation to the rest of us?
I was also heavily involved in building a Discord bot and a Telegram bot for using the Ray App Studio — which was frankly amazing — and built a lot of internal tools along the way. The bots I built from scratch responded to chat messages and images and baked apps via the Ray App Studio, piping content from one place to another through API calls — designed, learned, and implemented in a matter of days.
I wrote the Gherkin tests for manual testing first, then converted them to pyautogui — keeping manual and automated testing cohesive, so the whole app stays manually testable and most of it is automatable through a screen-driving controller. That controller became Raybotics — a pyautogui controller for gtest that ran the test suite via CI/CD on a Windows install (and locally). The Python scripts detected specific elements on the browser's app screen, clicked them, verified they changed on a second click, and confirmed states were captured fully — catching GUI regressions: an element that moved, stopped working, or broke when a new icon or a new system landed. Results were piped to Discord, so failures were visible immediately to the whole team.
I consider the Raymo / Rayma system one of the highlights of my career: a standalone Claude Code instance — effectively a developer that accumulated skills — driven from Discord. What it could do:
This is the tester-attacking-the-developer-angle taken to its conclusion: automation, prompting, and git grep turned into a shippable pipeline. GitHub Actions were used heavily across the role.
Across the role: manual and UX testing of the Chromium-based agentic AI browser (macOS and Windows); focus on usability, workflow optimisation, intuitive keyboard shortcuts, Picture-in-Picture, and agentic-browsing flows; experiments on embeddings, encodings, knowledge/hypergraphs for better recommendations; memory-system architecture for long-running sessions; evaluating pull requests, suggesting features, and working directly with developers and the product owner.