Deep Nexus was founded in 2017 as a research and development firm focused on identifying repeating structure in complex, noisy environments, beginning with financial market data and applications in quantitative trading systems.

We initially explored neural networks and, by mid-2018, developed deep learning models integrated into a full production stack for automated algorithmic trading. In January 2020, we expanded into portfolio-level modeling to capture cross-asset relationships and scale across large universes of instruments. While these systems often exceeded performance expectations, our research consistently revealed fundamental limitations in conventional machine learning related to non-stationarity, noise, and sparsity as well as an inability to reason about structure explicitly.

Addressing these limitations led to the development of a proprietary framework grounded in algebra and information theory. By early 2025, this work resulted in a deterministic, representation-first approach that replaces learned embeddings with explicit structure and geometry. It became clear this framework was not limited to financial markets; the same principles apply to any domain defined by discrete structure and combinatorial complexity.