We describe the methods and results of our submission to the 11th Social Media Mining for Health Research and Applications (SMM4H) 2026 shared Task 2. The task focused on the automated detection of insomnia using real-world clinical notes from the MIMIC-III database. Subtask 1 involved the binary classification of a patient's overall insomnia status, while Subtask 2 required multi-label classification and character-level evidence span extraction for specific insomnia-related criteria. We employed both encoder-based transformer models and decoder-based LLMs (Vaswani et al., 2017). For character-level span evidence extraction, we coupled the model's predictions with a rule based system.