A forecasting system that predicts utility consumption for each part of a manufacturing operation and publishes the results to a self-service web portal.
Illustrative forecast with 80% / 95% intervals · not real data
This is industry work described in general terms. Employer names, data, and screenshots are intentionally left out, and the chart above is illustrative.
01The problem
Energy and utility budgets were built from spreadsheets and rules of thumb. Each segment of the operation behaves differently, so one model never fit them all.
02What I built
Automated data pipelines that assemble consumption history and drivers on a schedule
Linear regression, random forest, and XGBoost models evaluated with 10-fold cross-validation
Automatic per-segment model selection: each segment gets whichever model forecasts it best
A reusable MLOps framework that validates about 10 candidate models each run and promotes the winner to production
An R backend with REST APIs and an interactive Shiny portal, so planners can explore forecasts and ranges themselves
03The result
Planners get forecasts with uncertainty ranges instead of single guesses
Models retrain and re-validate without manual work, and only promote when they beat the incumbent
Need something like this?
Tell me about the problem. I reply within one business day.