About

Hubert Misonia is a graduate researcher at the Centro de Investigación en Computación (CIC-IPN) in Mexico City, affiliated with the Modeling & Simulation Laboratory. His current work concerns the application of graph neural networks to the spatial reconstruction of extensive air showers (EAS) in ground-based gamma-ray observatories, specifically within the context of the Southern Wide-field Gamma-ray Observatory (SWGO). The thesis develops BiLayerEdgeConv, a dual-branch GNN architecture that processes the electromagnetic and muonic layers of a double-layer water Cherenkov detector array as separate graph structures, fusing their representations through bidirectional multi-head cross-attention with learned gating coefficients.

Prior to entering academic research, Misonia spent seven years as a software engineer at Kadea Software in Kinshasa, Democratic Republic of the Congo. There he led the development of hospital management systems handling patient records, appointment scheduling, and billing across multiple facilities; legal document platforms with granular role-based access control and audit trails; and school administration tools deployed to institutions across the city. The technology stack ranged from React and Node.js to .NET and PostgreSQL, with engineering teams of up to eight members.

The BiLayerEdgeConv architecture applies dynamic edge convolution (EdgeConv) operators to extract local geometric features from k-nearest-neighbor graphs constructed over detector station coordinates. Each branch processes one detector layer independently — the upper branch handles electromagnetic signals (charge, timing, PMT amplitude), while the lower branch processes muonic responses from the shielded detector volume. Branch outputs are merged via a bidirectional cross-attention mechanism in which each layer's node embeddings attend to the other layer's spatial context, capturing inter-layer correlations that single-branch models miss. This fusion strategy yields a 20% reduction in mean absolute error relative to the Centre-of-Gravity analytic baseline on CORSIKA/Geant4-simulated shower data.

Originally from the Democratic Republic of the Congo, Misonia speaks French and Swahili natively, Lingala at an advanced level, and is proficient in English and Spanish. His international trajectory — from Kinshasa to Mexico City — reflects the collaborative nature of large-scale astroparticle physics, where detector arrays span continents and working groups operate across time zones and languages.

During his tenure at Kadea Software (2022–2024), Misonia served as lead developer on projects for clients including GoLegal (law firm financial management), Finasucre — Compagnie Sucrière de Kwilu-Ngongo (industrial audit implementation), USAID/FABS (wildlife crime awareness), and Entreprendre pour Apprendre (educational platform). He organized workshops on project management, quality assurance, and development security, and supervised teams of up to eight developers.