The online shopping sites have been registering phenomenal growth where colossal volumes of user-generated reviews are being created on a daily to a daily basis. These reviews are quite educative and offer great insight on customer perceptions, likes and experience on products. Such reviews are nevertheless typically written in multi- and code-mixed language, so they are a huge challenge particularly in the linguistically diverse regions such as India. This paper presents the design and implementation of a multilingual sentiment analysis system that is able to address these issues. This system is founded on hierarchical pipeline which includes preprocessing of data, transliteration and translating. It normalizes and eliminates the input data, e.g., noise, e.g., emojis, abbreviations, irregular text patterns. Language and translation of all texts to a standard format are identified with the help of transformer-based models to analyze them. A transformer-based model is used to classify the text into a few types of sentiments, once normalized. The system also has visualization and reporting functions that provide data as charts, graphs, and summaries to determine trends and patterns in customer feedback. Overall, this application demonstrates that the combination of preprocessing techniques and existing deep learning systems can be helpful in processing multilingual and code-mixed data.