This paper demonstrates our system for shared task 4 of #SMM4H-HeaRD 2026 Workshop where a given doctor-patient dialogue is summarized into a clinical note in the corresponding SOAP format. Our proposed solution includes semi-supervised learning together with parameter efficient finetuning (PEFT) applied to a lightweight pre-trained QWEN3.5 model. Our model delivers competitive performance relative to its parameter count, and generalizes its performance to unseen test dataset.