Data Management for the Life Sciences

Anthropic just shipped Claude Science, an agentic research workbench for life sciences. The first thing we wanted to know was simple: point it at a real dataset through the Quilt MCP and see how far...

The reproducibility problem nobody fully solves alone

If you've ever tried to rerun an ML experiment from six months ago and ended up looking at four nested folders named v1_final, v1_final_REAL,

Most teams end up building an S3 data catalog one way or another. This is the 2026 playbook we run with biotech R&D teams: namespace design, schemas, workflows, lineage, search, governance, and the...

An NGS run produces hundreds of gigabytes that may be relevant to decisions two or three years later. This post is the working implementation pattern for tamper-evident NGS provenance on S3, with...

The technical posture for Part 11 on S3 is well understood. The documentation is where most pharma data teams slow down. This is a document-by-document walkthrough of the validation artifacts an...

Provenance for regulated data on S3 requires three primitives working together: cryptographic hashing and signing, S3 versioning with Object Lock, and immutable audit logs. This post walks through...

AWS offers more automation surface than any other cloud, which is both an advantage and a trap. These are the seven AWS tools we see actually working in biotech data platforms, ranked by the layer...

S3 versioning protects against accidents. It does not give datasets the schemas, lineage, or workflow checks that real governance requires. This is the end-to-end pattern we run with biotech teams...

21 CFR Part 11 was written for paper-era systems and applied to every electronic record since. This is a practical, vendor-neutral checklist for getting S3-backed pharma R&D records ready for a Part...

S3 is an excellent object store and an inadequate catalog. Teams that build past one bucket eventually need a versioned, searchable, governed layer on top. These are the ten capabilities we look for...