Pragmatic Frontier AI Lab

An applied AI lab, run by Vitalii Ratushnyi. We build the parts of an agent that the model does not ship with: memory, evaluation, and the harness around it.

Work

• Memory. Long-term memory for agents.
• Benchmarks. Task-level evals for model capability. We run them internally first and open-source the ones that keep predicting.
• Harnesses. Skills, tools, and pipelines that make an existing model finish more of the job unattended.
• Generative UI. Interfaces developed by the model, for the model.

Contact

v.ratyshnij@gmail.com

vitalii.connectingdots.fyi