Research and engineering groups I work within — one academic, one open-source.


Intelligent Computer Systems Research Institute

Research Fellow University of Miami

Approaching its 40th anniversary, ICSRI assembles researchers across machine learning, neuro-symbolic AI, platforms, personalized medicine, and AI ethics, advancing theoretical and computational work while probing the behavioral dynamics of how people interact with intelligent systems. As a center of excellence within the University of Miami Herbert Business School, it conducts and disseminates leading-edge research, advises industry and policymakers, and convenes conferences and workshops for the wider community.

My work there sits on the machine-learning-for-economics side: neural state-space architectures for non-stationary time series, and the interpretability question of what a learned spectral decomposition actually tells you about a regime change. The NS-SDN line of papers is the primary output.

Institute page

Output

Work Type Status
Non-Stationary Spectral Decomposition Network: Adaptive Spectral Emission Heads and Frequency Modulation Paper Under review
Non-Stationary Spectral Decomposition Network for Econometric Time Series Forecasting Paper FLAIRS-39, 2026
ns-sdn — neural architecture for decomposing non-stationary time series into interpretable trend and spectral components Software In development

toros-dev

Founder and Lead Developer toros-dev on GitHub

toros-dev is a research-oriented engineering group building high-quality, open-source infrastructure for the acquisition, transformation, and analysis of financial disclosure data, with an emphasis on reliable, reproducible, model-ready systems spanning structured extraction, XBRL parsing, and the reconstruction of consistent financial time series across filings, companies, and reporting regimes.

Its flagship project, toros, is a modular Python toolkit that represents complex financial objects as first-class typed structures behind a clean, extensible dataframe interface — emphasizing deterministic pipelines, robust API and SDK design, cross-filing normalization and entity resolution, and integration with modern data workflows including pandas, polars, and dask. Surrounding it is a set of source-specific acquisition libraries, beginning with edgar-sec, a client for the SEC EDGAR API that handles retrieval, rate limiting, and parsing.

My role is founder and lead developer: I set the architecture, own the public API surface across the stack, and maintain the release and distribution pipeline. The group’s philosophy is grounded in research-grade software engineering — combining econometrics, machine learning, and systems design to produce tools suitable for both academic research and production analytics.

Organization

Output

Project Role in the stack Status
toros — DataFrame extension for representing complex financial objects Representation In development
edgar-sec — Python client for the SEC EDGAR API Acquisition Released

Full project detail →