Deepan Jayaraman
Ph.D., Indian Institute of Technology Madras
Data-driven decision making — how choices should be made when the evidence is thin, the consequences are asymmetric, and the model informing the choice is itself estimated.
I am a quantitative researcher and a working data scientist. My doctoral work at IIT Madras developed an L-moment framework for uncertainty quantification, Bayesian probabilistic risk assessment and robust design optimisation under scarce data containing extremes, published across three Q1 journals. Four years in industry since — at Caterpillar and Onward Technologies — have put those methods into production: generative AI and agent systems, predictive modelling, machine-learning-enhanced digital twins and decision-support systems. The two feed each other. Production supplies the problems worth solving, and the methods have to hold up once a real decision depends on them.
One question, several domains
Data-driven decision making is the spine of everything I do. The methods travel; the domains change.
Methods against domains
The same claim, as evidence. Rows are methods, columns are domains; a filled cell means that method produced work in that domain.
Research in brief
Uncertainty quantification from scarce, extreme-laden data
L-moments provide robust statistical characterisation where product-moment estimators become unstable, yielding reliable estimates from limited samples containing extremes.
Bayesian probabilistic risk assessment
Integrating Bayesian inference with L-moments, using distribution-independent approximations for conditional probabilities to widen applicability under uncertainty.
Prescriptive analytics and the model-to-decision handover
When an optimisation program's objective is a learned model rather than a specified one, the search inherits whatever that model attributes — and explanation has to govern the passage from model to decision.
Industry in brief
Generative AI and agent systems
Python-based generative AI tooling and agent workflows for analytics use cases across engineering solutions — my current focus at Caterpillar.
Digital twins and predictive modelling
Machine-learning-enhanced digital twins for strategic planning, including battery thermal and ageing management, alongside time-series models for engineering and business planning.
Data engineering and decision support
ETL pipelines, model selection for the dataset and the question at hand, and interactive dashboards that support engineering decisions rather than report on them afterwards.
Experience
-
Jul 2023 – present
Senior Associate Engineer — Data Science
Caterpillar India Engineering Solutions, Chennai -
Jun 2021 – Jul 2023
Data Scientist
Onward Technologies Ltd., Chennai -
Aug 2015 – May 2021
Research Assistant
Department of Engineering Design, IIT Madras -
Aug 2013 – Apr 2015
Lecturer
St. Peter's University, Chennai