Research

Agent systems

We design agents for long-running work: memory, orchestration, task scaffolds, and evaluation. The central question is not only what an agent can do, but how consistently it can do it.

Cost, accuracy, and execution time are treated as design constraints. We measure them, make trade-offs explicit, and improve the system through repeatable experiments.

Research infrastructure

Complex research needs more than a final answer. We build systems that preserve evidence, assumptions, and review steps so a result can be examined and used again.

Our work connects literature and data with analysis, keeping the reasoning behind a conclusion accessible to the people who need to act on it.

Domain simulation

We model domain knowledge and data to compare scenarios before a decision is made. The aim is practical: create a clear space to test alternatives, uncertainty, and constraints.

Current work spans bio, medicine, and drug discovery; scientific and engineering simulation, including CFD; finance and quant; economics; business; and policy.

Hylē

Hylē is REFRESH’s first research and simulation platform. It is where we develop models and systems for investigating complex questions and comparing possible futures.