Our work, in the open.
Papers, whitepapers, and case studies from Rainwater Labs — alongside the third-party research that shapes how we think. Everything we publish is open by default.
DEW: Reclaiming Classroom Thinking in the AI Era
The mission brief and full technical architecture behind our Socratic foundation model for education — how the method is baked into the weights, why we refuse to hand students finished answers, and what that does to long-term retention.
Logits-Level Guardrails for Pediatric AI
We show how DPO combined with a runtime LogitsProcessor can lock the Socratic process and forbidden topics into a model at the weight level — making unsafe outputs physically inexpressible rather than filtered after the fact.
Training Frontier Models on a Fraction of the Energy
A study of the architectural and data choices that let us cut energy per token by an order of magnitude without sacrificing downstream capability — and the trade-offs we accepted to get there.
Sovereign AI: Why Where Your Model Runs Matters
A practical framework for thinking about data residency, open weights, and on-device deployment — and why dependence on AI controlled by a few is a greater risk than the capability itself.
The Economics of Affordable Inference
Cost-per-token benchmarks against incumbent providers, the unit economics of running lean models at scale, and a clear-eyed look at what it takes to make frontier capability genuinely affordable.
New papers, in your inbox.
A short note when we publish something new — research, whitepapers, and the outside work worth reading. No noise.