Deep Nexus was founded in 2017 to investigate how computational systems can extract useful structure from complex, continuously changing data.
What began as research into deep learning and time-series prediction gradually became a broader investigation into information, representation, state-space, and control. Over nearly a decade, that work moved from predicting what a complex system might do next to determining the state of a system, understand the space of states available to it, and navigating toward a desired outcome.
That progression ultimately led beyond financial markets and into physical systems and to the development of our STRATA platform.
2017 — Deep Nexus Begins
Deep Nexus began with research into deep neural networks for sequential and time-series data, initially focused on sequence-to-sequence architectures. Financial markets provided a demanding experimental environment with large quantities of real-time data, continuously changing relationships, substantial noise, and an objective measure of whether a model was useful. By August 2017, Deep Nexus had completed its first predictive models, with internal benchmark testing on foreign-exchange and blood-glucose time-series data producing results that exceeded selected published models.
2018 — From Models to Live Systems
Research moved from offline prediction to real-time operation. Deep Nexus developed its first live financial-market system integrating streaming market data, TensorFlow inference, broker connectivity, and automated order-management logic. Financial markets demonstrated what would become increasingly important; a model does not operate on a static dataset and it operates inside a system whose state continues to change and can adapt itself to the model’s outputs.
2019–2020 — Beyond Prediction
As the research progressed, Deep Nexus began investigating information theory and control theory as alternative ways of thinking about dynamic systems.
By 2020, the work was increasingly moving away from fixed-horizon questions such as “What will this variable be at some future time?” toward continuous state estimation: “What is the system’s state now, and how should an action change as that state evolves?”
This marked an important conceptual shift. Prediction asks what happens next. Control begins by asking where the system is now which is especially useful in a noisy system.
2021–2024 — The Representation Problem
The shift toward state created another challenge: how should the state of a complex, non-stationary system actually be represented?
Learned representations can be powerful, but they can also be opaque, unstable, and dependent on the data used to fit them. Deep Nexus increasingly focused on explicit representations in which the information and relationships defining a state could be preserved and examined directly.
By 2024, this had become a central research program with the development of state-space representations intended to address noise, non-stationarity, dimensionality, and sparsity without relying exclusively on learned latent representations.
2025 — An Information-Theoretic Framework
In early 2025, this work culminated in our Algebraic Information-Theoretic framework, initially applied to options markets.
The approach was representation-first. Rather than beginning with a model trained to predict an output, it sought to make the information and relationships defining system states mathematically explicit.
This approach raised a much larger question: could the same principles be applied to other complex systems? Deep Nexus began extending these ideas to materials.
Late 2025 — From Information to Matter
Materials presented a fundamentally different challenge. Financial states exist computationally as observations; materials exist physically and are constrained by chemistry, thermodynamics, kinetics, phase behavior, and processing history.
Deep Nexus began investigating how matter could be represented computationally while preserving explicit information about composition and relationships among material states.
But this work also exposed a limitation of purely computational materials discovery. A mathematical representation of a possible state is not the same as a physical pathway for reaching it.
2026 — From Mapping to Navigation
In early 2026, Deep Nexus began asking whether its work in information theory, state representation, and control could be coupled directly to physical hardware.
Instead of using computation only to identify potentially interesting material states, could a system measure its current chemical state, apply a controlled intervention, observe the resulting transition, and determine what to do next? Chemical processing could then be considered not merely as a fixed recipe, but as a trajectory through chemical state-space. This became the foundation of our STRATA platform.
