Very interesting! Could you elaborate a bit on the parallels with Keras, other than the API similarities? What aspects of working with LLMs does SynaLinks make easier?