Subtree

Subtree Inc., which later rebranded to Dotscience, was a UK-based startup founded in 2017 by Luke Marsden, a recognized veteran in container storage. The company launched with a significant $10 million seed investment led by DDN Storage. Marsden's journey into founding Subtree was built on extensive experience in infrastructure software and distributed systems. Before Subtree, he was the founding CEO of ClusterHQ, where he created the open-source container data volume manager, Flocker. His deep involvement with the container ecosystem, including work with Docker and Kubernetes, and a role as Head of Developer Experience at Weaveworks, directly shaped the company's mission. He identified a critical gap in the DevOps workflow: the lack of version control and reproducibility for data, similar to what Git provides for code.

Subtree's initial offering consisted of two core products: dotmesh and dothub. Dotmesh was an open-source utility designed to function like "Git for data," allowing developers to capture, version, and share the entire state of a containerized application—including multiple databases and files—in a single atomic unit called a "datado. Dothub served as a centralized cloud repository, akin to GitHub, for storing and collaborating on these datadots. The business model included a free tier for dothub, with paid subscriptions for increased storage and team features.

The company pivoted to focus specifically on the emerging Machine Learning Operations (MLOps) market, rebranding as Dotscience. The core technology was adapted to address the unique challenges of the AI/ML lifecycle. The Dotscience platform provided data scientists and ML engineers with tools for reproducibility, collaboration, and governance. It allowed teams to version control not just code, but also the data, models, and parameters used in experiments, ensuring full provenance and traceability from a production model back to its origins. The platform integrated with existing tools like Jupyter notebooks and GitLab, aiming to bring DevOps best practices to AI development without forcing teams into a restrictive ecosystem. It catered to a growing market of enterprises in sectors like finance and autonomous vehicles investing heavily in AI. Despite securing clients and developing its platform, Dotscience announced it was shutting down in May 2020, citing an inability to secure further funding amidst the global economic situation.

Keywords: MLOps, data versioning, dotmesh, DevOps for ML, Luke Marsden, container data management, model reproducibility, data provenance, AI lifecycle management, ClusterHQ, Kubernetes storage, Git for data, datadots, collaborative data science, machine learning platform, model deployment, TrueLayer, DDN Storage, Subtree Inc, Flocker, data lineage, experiment tracking, model governance, CI/CD for machine learning