The escalating burden of neglected tropical diseases (NTDs) in East Africa, particularly schistosomiasis and lymphatic filariasis, is compounded by emerging drug resistance and a paucity of novel therapeutic candidates. This study aimed to identify potential small-molecule inhibitors against validated NTD targets using a machine learning-driven virtual screening pipeline. A two-stage computational framework was developed, integrating a graph neural network (GNN) trained on the ChEMBL30 bioactivity database (n = 1.8 million compounds) with a Bayesian optimisation algorithm for molecular docking against Schistosoma mansoni thioredoxin glutathione reductase (TGR) and Brugia malayi asparaginyl-tRNA synthetase (AsnRS). The model achieved a receiver operating characteristic area under the curve of 0.94 on a held-out test set. The top 0.5% of 10 million screened compounds (n = 50,000) were docked, yielding 127 high-affinity candidates (binding energy ≤ −9.5 kcal/mol) with favourable ADMET profiles. The novelty lies in the application of a multi-task GNN architecture that jointly predicts target affinity and selectivity, reducing false-positive rates by 18% compared to single-task baselines. The logistic regression model for hit prioritisation was specified as $\log(p/(1-p)) = \beta_0 + \beta_1 \cdot \text{GNN score} + \beta_2 \cdot \text{docking score} + \epsilon$, with 95% confidence intervals for $\beta_1$ and $\beta_2$ excluding zero (p < 0.01). These findings provide a validated computational workflow that can be deployed for rapid NTD drug repurposing in resource-limited settings, though in vitro validation remains necessary before clinical translation.