Much of the scientific software workflow is still imperative, requiring scientists to repeatedly write complex and error-prone algorithms.
However, over the past decade software engineering has moved to declarative tools—Kubernetes, Docker, React, Terraform—and reaped huge benefits. Science should do the same. Declarative science would make experiments more reproducible and researchers more productive. It's a foundation for better tools, the kind that let scientists focus on ideas instead of implementation details.
Cast, parameter-driven instantiation of complex objects.
Labfile, a file format for declarative orchestration of scientific experiments.
Lab, an interactive application for managing a Labfile and scientific assets.
Anot, a simple tool for extracting structured data from source code comments.
Vyper, an extended Python syntax introducing scientific units, probabilistic variables, hyperparameter management, and other features.
Computational scientists spend a lot of time on rote tasks, rather than asking questions. With the advent of LLMs, we can build tools that speed up the process of interpreting existing science and ultimately answering new questions.
Method, a tool using Anot and git to automatically track code changes via the scientific method.
Puppet, a tool that generates a candidate codebase for a given research paper, to aid reproducibility.
Editor, an AI-native app for integrating scientific knowledge into the writing process.