refactor: Task3 reward model changed, agent adjusted for new model

refactor: Improved grading logic for task 2

Structure Changed, files reviewed