The world today is saturated by digital documents, especially PDFs, and trying to find some information in one feel like looking for a needle in a haystack. The traditional method of searching through keyword matches does not reach the mark because these just do not grasp the meaning beyond our questions, that is why we came up with this smart PDF chatbot that can sense and answer questions pertinent to its document content genuinely. Our system is an experiment with lots of methods of teaching the computer to understand text from simple approaches like Word2Vec. The system involved extracting the text from pdf document and processing it to form embeddings and thereafter storing them into database and finally generating a valid response based on embeddings of query, that is similar to how a very well-trained specialist might provide you with more precise answers but would be more expensive to consult. Our comprehensive analysis showed that transformer-based models (Google Generative AI embeddings) achieved superior accuracy over traditional algorithms like N-grams, TF-IDF, and BoW. Our work thus does not only reveal what the current embedding approaches can do but also compares traditional embedding approaches on the basis of their memory efficiency, time complexity, and similarity scores.