Personally Identifiable Information (PII) identification, redaction, and de-classifying if necessary is of prime importance not just to meet several compliance standards for healthcare, banking, insurance, and other business lines but also to safeguard and protect consumer interest. There exist methods to extract PII from textual corpus but for images there is scant of easily trainable and deployable models. In this paper we have explored the use of transfer learning combined with dilated convolution in a Convolution Neural Network (CNN) to find the segments of an image which corresponds to a PII looking field and demarcate it. The segment then identified is classified in a different color apart from a body color map that is pre-configured for our images. The training set consists of original images and masked segments that represent PII. The test set consists of newly obtained and invoice images that contains PII information in same spatial coordinates. The network built also contains up sampling to modify the convolution lenses that are used to visualize the field of interest of the image i.e., the segments so that the field of vision is broadened. Serverless Architecture is an emerging trend in modern cloud computing that is becoming increasingly popular due to its flexibility, scalability, cost efficiency, and agility. It allows organizations to focus on their core business objectives and development rather than managing their own servers. This paper focuses on the use of Microsoft Azure Functions [10] and Continuous Integration and Continuous Deployment (CICD) Pipelines for deploying serverless architectures. It discusses the benefits of this approach, including cost savings, scalability, and agility. The paper also examines the challenges associated with this approach, such as security and cost, as well as how to overcome these challenges