Science Data Infrastructure Platform

Computational scientists lack modern devtools for managing experiments, data provenance, and collaboration. Build a platform offering declarative experiment tracking, data versioning, provenance logging, and secure data sharing tailored for research teams. Target: PhD students and research labs doing computational work.

MLflow, DVC, and Weights & Biases already own significant mindshare in the ML experiment tracking space, and the broader scientific computing world has seen rising adoption of workflow tools like Snakemake and Nextflow — meaning the pain is real but partial solutions exist. The distinction worth pursuing is provenance and collaboration for non-ML computational science: genomics, climate modeling, physics simulations, where W&B is irrelevant and the tooling is still bash scripts and shared filesystems. The $5k–$20k/mo revenue band is plausible only if you can land institutional or lab-level contracts, since individual PhD students have no budget — that means a longer sales cycle and dependence on grant renewal cycles, which compresses growth. The most likely failure mode is that researchers tolerate bad tooling indefinitely rather than adopt new software, especially if it requires changing existing pipeline code; adoption inertia in academic computing is severe and historically underestimated by founders coming from industry.