Recent work proposed a process algebra for machine learning, termed the Transfer Meta Process Algebra for Learning (TMPAL), wherein formal, abstract learning systems are the objects of the algebra and transfer learning and meta-learning are its operators. This paper presents an approach for implementing TMPAL using the Meta Type Talk (MeTTa) programming language, termed MeTTa-TMPAL. Each learning system is treated as a MeTTa “atomspace” and has well-typed inputs, outputs, and response functions (an algorithm and hypothesis). Operators are defined as functions on learning systems with by-construction guarantees that enforce key constraints of TMPAL using MeTTa’s rewriting capabilities. Moreover, operators can be defined as learning systems themselves, harnessing those rewriting capabilities towards self-writing learning processes. A brief discussion of stability constraints for self-writing processes is included. Open questions remain regarding additional constraints for learning operators.