Open-Source Inventions

In 2022, Deepak introduced the Generative Value at Risk (GVaR) framework, a probabilistic generative ensemble approach to financial risk measurement.

GVaR integrates established mechanisms underlying historical, parametric, and Monte Carlo methods within a probabilistic machine learning framework. Most importantly, it provides an interpretable and auditable method of risk measurement.

The posterior predictive distribution that generates future returns is the fundamental risk object in the framework. VaR, Expected Shortfall (ES), and other tail-risk measures are then derived from this distribution, providing a transparent link between model assumptions, parameters, historical data, uncertainties, simulated outcomes, and risk measures.

Checkout the Python code in our Github repo here. See also Deepak's presentation on GVaR framework given at the 2023 O'Reilly Media conference on Generative AI in Finance.

In 2005, long before machine learning became an industry buzzword, Deepak developed a probabilistic machine learning method and software system for managing the risks and returns of project portfolios. See his patent filing.

The system learned from data in time sheets and work‑completion percentages to continually update the probabilities that a project would remain within its time, budget, schedule, and resource constraints. The estimated probabilities were then aggregated at the program and portfolio levels.

It is a unique probabilistic framework that has since been cited in patents filed by IBM, Fujitsu, Accenture, Huazhong University of Science and Technology, State Grid Zhejiang Electric Power Co Ltd, Zhejiang Gongshang University, Kyung Hee University's Industry-Academia Collaboration Foundation.