Create an end-to-end framework that turns natural language feedback into model weight updates. This enables LMs to learn from user interaction more efficiently than current fine-tuning pipelines.