Resilient DevSecOps: leveraging large language models and chaos engineering for automated threat hypothesis validation Tesis