In this paper, we present a context-aware advertisement recommendation engine for web pages. Context of the web page is obtained by summarizing the text in terms of keyphrases that are relevant. These key phrases are automatically extracted by various transformer architectures, namely the keyword bidirectional auto-regressive transformer (KeyBART) and sentence bidirectional encoder transformer (SBERT). The focus of the paper is to increase the key phrases that are relevant by fine-tuning the KeyBART architecture as it increases the chances of a higher similarity score that is computed by the SBERT architecture. The summarization evaluation is performed by the ROUGE score, and the autotrained and finetuned models of the KeyBART perform equally better than the KeyBART base models, signifying the effectiveness of fine-tuning the models. The advertisement is recommended based on a empirical threshold that is computed based on the similarity that the SBERT engine provides by computing the pre-stored advertisement key phrases and the generated key phrases.