About a year into my PhD, I began noticing a gap in science communication. There was a growing interest in general-audience scientific media, distilling recent cutting-edge research into digestible content on social platforms. I’m an enormous fan of this outlet for non-technical audiences, but it’s often overlooked that technical audiences face the same problem. When a new discovery is published, most of the effort in an article goes toward rationalizing an intellectual contribution, while reproducibility is secondary, and ease of implementation is an afterthought. Because of this, most discoveries take a long time to digest and implement by practitioners but getting new technology in the hands of practitioners is where real impacts are made.
Contextualized.ML began as an experiment to fill this gap and get my research to work out-of-the-box for practitioners and instill knowledge as fast as possible. Like most open-source software, it would be modular, tested, and documented in several levels of complexity to cater to a wide range of technical audiences. To clarify intellectual contributions, it would ship with tutorials on the more complex topics.
What began as an isolated open-and-shut experiment quickly outgrew my individual ownership and is now a thriving community-driven project. Researchers from around the world contribute to Contextualized.ML because they find the work enjoyable and interesting, and they have been able to implement and improve it without needing months to replicate a pilot study. The development of this software has reflected my development as a scientist, and in both regards I’m far from finished – but always learning and always improving.