zlacker

With LLM itself using those graphs to validate its answers, etc. And at some point it will be internalized into architecture as a graph attention layer.