[IJCNN 2021] Unified Spatio-Temporal modeling for traffic forecasting using Graph Convolutional Network

A Task-Oriented graph SAmpler for HGNN

CLEVR graph: A dataset for graph based reasoning

End-to-end graph AML investigator with explainable scoring, what-if analysis, and operator-facing UI.

Graph-RAG for Customer Journey Intelligence using NetworkX + LLM. Path-aware retrieval outperforming vector RAG on temporal queries, cohort comparison with real statistics, 5 pre-built analytics queries, and fully dockerized FastAPI/Streamlit architecture deployed on HuggingFace Spaces.

A deep learning architecture combining spectral graph neural networks with curriculum learning for HOMO-LUMO gap prediction on PCQM4Mv2. Features a dual-view architecture with Chebyshev polynomial-based spectral convolutions and complexity-driven training schedules.

Graph-native fraud ring detection with hybrid ML/DL scoring, explainability, and production observability.

A deep learning approach for molecular property prediction that introduces hierarchical attention pooling to capture scaffold-aware representations. The model aggregates atom features within functional groups before global pooling, combined with scaffold-based curriculum learning for improved generalization across diverse chemical structures.

Graph representation learning — reproducing and analyzing core methods for academic study

Self-Supervised Similarity Learning of Floor Layouts