Define the package
Identify the paper, dataset, notebooks, intermediate artifacts, lab context, and access limits.
Scientific data infrastructure
New Atlantic
We are laying the data foundation for scientific research agents. AI models need richer, better-contextualized research data to become useful to practicing scientists, but most high-quality scientific data is trapped under institutional complexity. We partner with universities and academic labs to transform this data into packages for training scientific agents, while helping academic labs structure their data for future research.
What we do
We help labs preserve context, structure data, and create future upside.
The most valuable scientific data is rarely just a cleaned table. It lives across lab notebooks, intermediate results, negative results, analysis code, instruments, and the judgment behind a published paper. New Atlantic works with originating labs and universities to turn that context into a well-described research asset that can support future scientific agents.
For PIs and labs
Your lab data can become durable research infrastructure.
We help labs identify the data and context around a strong paper, organize it into a coherent package, and translate it into tasks that scientific agents can learn from or be evaluated on. The goal is to preserve the work in a form that is useful beyond the original publication while keeping the PI and originating lab connected to any downstream value.
- Structure datasets, notebooks, code, intermediate results, and negative results
- Capture the tacit context that makes the data scientifically meaningful
- Build around work the lab already understands deeply
- Create a revenue-sharing path for the university and originating lab
For universities and TTOs
A staged way to make research data useful without losing control.
We start with a narrow non-commercial pilot so the university can define the data package, access boundaries, and review process before negotiating commercial terms. New Atlantic handles packaging, environment design, and buyer readiness, while the institution and lab retain a clear role in permissions, licensing, and revenue share.
Scoped pilot
define the data, access boundaries, and review process before broader rights are considered
Shared upside
keep universities and originating labs connected to downstream licensing value
TTO-ready path
route frontier-lab demand through diligence materials and institution-ready agreements
Commercial path
Academic data becomes licensable infrastructure for scientific AI.
Frontier AI labs need training and evaluation environments that reflect real scientific work, not just public benchmarks. New Atlantic turns selected lab packages into legally licensable, agent-ready environments with stable tasks, feedback, held-out splits, and reward-integrity protections.
- Universities avoid one-off negotiations with every AI lab
- PIs gain a practical route for their data to shape scientific AI
- Commercial upside is shared instead of extracted from the lab context
Pilot model
Designed to move from lab context to shared research infrastructure.
Structure the asset
Organize the materials into agent tasks with documentation, feedback signals, and evaluation splits.
License if ready
Convert successful pilots into commercial environments with university and lab revenue share.
What gets built
Each package keeps scientific context attached to computable tasks.
Data descriptions and access boundaries
Paper-linked tasks and verification signals
Containerized setup and grading hooks
Named numeric subscores
Train, test, and transfer splits
Baseline, golden-path, and integrity checks
Start a conversation