Projects 

AquiferPulse: Groundwater early-warning (in preparation)

AquiferPulse is an early-stage groundwater early-warning prototype for Senegal, designed to integrate rainfall, satellite-derived water-storage indicators, land data, and basin-level visualization to support monthly groundwater-risk monitoring. It is being built to produce a monthly view of national groundwater conditions, classifying each basin as alert, watch, or normal. 

Undergraduate Project

Supervisor: Prof. Sénie Tamba

This project focused on the hydraulic design of a drinking-water transmission system connecting Bargny, Diamniadio, Mbour, Pout, Mbissao, and Thiès in Senegal. We developed the route layout and longitudinal profiles, distributed the design flows across the different pipeline branches, selected commercial pipe diameters, and calculated velocities, head losses, available pressures, and energy lines for the network. The result showed how pipe sizing, elevation differences, and head losses affect pressure distribution along a water-supply system, with Thiès having the lowest remaining pressure due to its downstream position and higher elevation.

Supervised machine-learning drought detection at station scale (1994–2024) 

This project developed a supervised machine-learning workflow to detect monthly drought conditions across weather stations in Southeastern Australia using climate, vegetation, and soil-moisture indicators. The workflow combined SPI-3, SPI-12, NDVI, and root-zone soil moisture to create a drought-labeling system. The model development was carried out using a chronological train-test split, threshold calibration, and performance evaluation with ROC-AUC, F1-score, recall, and time-series validation.

The Random Forest classifier showed the strongest performance, achieving high drought-discrimination capacity on the test period. A transfer experiment to an unseen station showed that the model could rank drought risk but required recalibration before operational use. 

Spatial Generalization

The source-trained model was tested on an unseen station to evaluate spatial transferability. It showed good ranking capability, with ROC-AUC = 0.822, but weaker calibration for the new location, with F1 = 0.448. After light fine-tuning and retraining, performance improved, reaching accuracy = 0.856 and F1 = 0.662. 

Global AI Hackathon 2026 (Aspire Leadership Program x Cayu) 

Group project with Fathima. Illsan

24-hour global competition to design and conceptualize innovative AI solutions.