Government tender documents contain multiple labels associated with a single description that encompasses a substantial quality of vital information, and the precise extraction of pertinent codes is crucial for effective procurement procedures. The code extraction of government tenders is an issue that can be resolved through the Extreme Multilabel Text Classification (XMTC) task, in which each tender document is assigned the most pertinent labels from a large label collection. The Prediction for Enormous and Correlated Output Spaces (PECOS) framework of XMTC has been an efficient solution for code extraction of government tenders. Initially, the Semantic Indexing technique of the PECOS framework efficiently retrieves candidate codes by structuring the vast label space into a semantic index. Utilizing linear or neural matches of machine learning matching techniques, semantic indexing reduces the search space. Recursive Transformer models of XMTC provide better accuracy at the expense of longer training times, and X-Transformer has been used to enhance the labels and features for better results. In this paper, the ranking and prioritizing of True Positives or True Negatives has been conducted through PECOS, which results in the generation of the most pertinent document. Through PECOS modular components, the XMTC issues such as accuracy, computational efficiency, and correlations between labels in diverse sparsity can be efficiently handled. The results have proven that the proposed approach achieves high accuracy, which improves the finding of tenders from different resources by 90%. Its adaptability enables practitioners to alter algorithms for every stage according to government tender requirements. PECOS also performs fast inference, which makes it appropriate for real-time situations.