Gaslytics
Leading supply and storage development for a natural gas modeling system.
Gaslytics is a comprehensive modeling system covering daily North American natural gas supply, demand, storage, and pipeline flows. The system generates both historical data and forecasts at granular geographic levels. I was a lead developer, leading development of the supply and storage models. I also contributed to demand models built with linear regression and MLflow, helped architect how the models flow together, and deployed models to production as business requirements evolved.
01 / The Challenge
The supply, storage, and demand models needed to work together within a production modeling system focused on historical data, with some forecasting. Their individual algorithms and pipelines also had to adapt to continually changing business requirements.
The work required both model development and architectural decisions about how the models connect, followed by deployment into production.
02 / The Approach
- Led development of the supply and storage models, working through their algorithms and data pipelines as requirements changed.
- Contributed to the demand models, which use linear regression and are set up with MLflow.
- Helped architect how the models flow together across the platform.
- Carried the modeling work through to production deployment.
03 / The Impact
- Delivered supply and storage model development as a lead developer on Gaslytics.
- Contributed to the demand modeling system and the architecture connecting the models.
- Deployed models to production while adapting the implementation to evolving business needs.
Related production responsibility
Operating the legacy natural gas modeling system
Alongside developing Gaslytics, I independently maintained the legacy MATLAB natural gas modeling system that it replaced for approximately three years. This system supported the former Bentek business’s natural gas models, reports, and analysis.
I owned maintenance, performance, and operations, including monthly model performance analysis, retraining, and deployment. I also added finer pipeline-flow granularity between New Mexico and Texas and modeling features for new LNG export facilities.