Experience

Built a cost-forecasting layer for the seawater reverse-osmosis desalination unit at OCP Phosboucraa. The site's dashboard only reported a week's operating costs after the fact; I rebuilt it and added a predictive layer that estimates the coming week's cost along with its uncertainty.

Engineered a weekly dataset from raw SCADA exports and focused the modelling on the specific energy consumption of the main stream, the dominant cost item. A leave-one-out Ridge regression served as a diagnostic, a hierarchical Bayesian model in PyMC gives the predictive distribution, and a Monte Carlo simulation combines all seven cost categories into a distribution of the total weekly cost.

Added a conversational assistant that answers questions about past costs and forecasts in plain language, using LLM tool calling over the dashboard's own data, with a local Ollama model as a drop-in option for confidentiality.

Built FESURBAN, a full-stack mobile app replacing a fully paper-based urban planning file system. Agents went from spending 15–45 minutes searching through physical archives to finding any file in under a second — a 5x productivity improvement confirmed in user testing.

Stack: Flutter frontend, Node.js/Express REST API, MySQL database with optimized multi-join queries across 8 tables, JWT authentication, and bcrypt security. Designed as a replicable digitization model for Urban Agencies across Morocco.