Graph Neural Networks
Second-order representations and structural kernels for graph learning.
overview
Our mission is to create useful, research-driven public tools with long-term value for scientific and societal benefit.
Our work bridges graph theory with neural network architectures — from process graphs and asynchronous computation to knowledge representation and agent memory systems.
Second-order representations and structural kernels for graph learning.
Graph engines and entity embeddings for structured AI reasoning.
Graph-native persistent memory for LLM agents via MCP.
Typed diff and evolution tracking for dynamic graph structures.
Event-driven neural networks and GPU task scheduling.
Computational trade-offs, causality, and neuroscience-inspired architectures.
Research-driven public tools built for long-term use. All repositories are open-source and actively maintained.
Graph-native persistent memory layer for LLM agents. MCP-compatible — Cursor, Claude, custom agents via FalkorDB.
Typed structural diff for knowledge graphs. Snapshot, compare, replay. HTML reports, cluster evolution.
Theoretical framework: exploration-exploitation trade-offs, pancomputationalism, neuroscience-inspired architectures.
Graph token injection into frozen LLMs. Benchmarked vs Text-RAG on MetaQA-3hop and WikiMovies.
New research-driven tools, neural graph representations, and asynchronous computing libraries are currently in development.