Hydrological forecasting / Predictive engineering
From sensor data to water-quality forecasts.
A predictive layer linking environmental time-series data, model evaluation, training and inference APIs.
The challenge
Water-quality measurements vary across time and locations. A useful forecast needs consistent time-series processing, evaluation suited to each parameter and a reliable path from model development to operational inference.
What we engineered
We delivered the predictive layer from data analysis and model development through training and an inference API. We compared Transformer and TimesNet approaches for hydrological time series and contributed to the data-management design around the deployment.
What changed
A custom deep-learning model deployed in production across tens of floating monitoring stations, producing parameter forecasts through the inference layer.
Scope & evidence
Production deployment and scale reported by ArtNovation in September 2026. The work covers a stochastic environment; quality is assessed through parameter-specific forecast errors.