Junctive Unstructured Learning for Incrementally Evolving Transformers
Juliet is an experimental AI framework built for incremental, evolutionary learning. Unlike static models that freeze after training, Juliet creates a living ecosystem of isomorphic entities (isos) — lightweight transformer-based learners that grow, adapt, and evolve over time.
Each iso maintains its own persistent memory (via vector stores) and ingests unstructured data from conversations, documents (.pdf, .txt, etc.), or other sources. Over time, isos fine-tune lightweight adapters on their unique history, giving rise to divergent traits, behaviors, and emergent intelligence.
Juliet goes beyond conventional RAG pipelines — it’s a wild evolutionary soup of transformers, where memory, state-space dynamics, and perturbations converge to produce creative and adaptive learning.