Infrastructure Risk Simulation Platform

PythonPrefectDaskAWSSageMakerAthenaS3SQLParquet

Note: Descriptions are abstracted to highlight transferable skills. No proprietary systems or data are included.

Project Overview

Developed a cloud-based simulation platform to support large-scale modeling of rare, high-impact risk events in critical infrastructure contexts. The system applies stochastic methods to simulate a wide range of potential outcomes across distributed assets, helping to assess systemic exposure and inform strategic decision-making. Millions of simulation outcomes are generated to assess potential impacts, presenting challenges in scalability and data throughput. The system described below addresses these challenges through distributed, parallel execution of simulations.

Scope

While I contributed to model development, the following sections focus on my work designing the system architecture and data pipelines that enable scalable simulation, ephemeral cloud compute for parallel jobs, and parametric analysis.

System Architecture

To support large-scale risk simulations efficiently, the platform dynamically provisions cloud compute resources based on demand. User-defined parameters sent through the API launch one or more simulation flows, such as for modeling hazard events or calculating their costs. An orchestration layer manages each flow, and workloads are executed in parallel to efficiently process millions of data points and handle I/O operations.

The platform also supports simultaneous execution of multiple simulation jobs, enabling parametric analysis. When users specify ranges for key parameters, the API automatically generates and launches a series of independent compute jobs, each receiving a unique parameter set. This capability is ideal for conducting sensitivity analysis and optimization.

Impact

Accelerated development and long-term adaptability by building a simulation framework with modular components, structured configuration, and automated testing.
Reduced simulation runtimes by hours through architectural changes that enabled parallel execution, optimized workload distribution, and improved I/O efficiency.
Enabled rapid parametric and sensitivity analysis by developing logic to dynamically provision ephemeral compute jobs with variable input parameters.
Improved developer experience with real-time monitoring of processing jobs, and streamlined workflows for interpreting and comparing simulation outputs.