sistemalabs: the foundation model factory

Transformers have succeeded in modeling language and beyond. We are working to industrialize the development of foundation models across domains.

⛓️ focus: states and actions

We train state/action foundation models. We interpret 'action' liberally across domains:

• people: clicks, purchases, transactions
• biology: cells, neural activity
• animals: behavior, communication
• systems: markets, robotics

🔧 tooling: sequifier

sequifier is our library for building foundation models.

It implements a unified workflow for single- or multivariate causal decoder-only transformer architectures and data preprocessing.

• configurable: Standard causal, autoregressive architecture with flexible specifications, input and target variables, and loss functions.
• scalable: Multi-GPU/Multi-Node training.
• ethos: Implement once. Use everywhere. Learn from each domain for the general case.

🏴 collaborations

We partner with organizations to build domain-specific foundation models—whether for internal deployment or open release.

• contract development: Hand over the data and requirements. We train the model end-to-end. You own the weights.
• enterprise: Embedded support for developing and deploying a sequifier foundation model.

Links

• learn more about us
• learn about our family of Ethereum transaction foundation models, EWE-1

© 2026 sistemalabs research OÜ.