<p>Traditional procurement processes in both public and enterprise sectors suffer from inefficiencies caused by unstructured requirement expressions, labour intensive document preparation, and the lack of standardized workflows. To address these limitations, this paper proposes an AI-driven integrated framework that automates procurement requirement understanding and tender document generation based on natural language processing (NLP) and knowledge graph (KG) technologies. The proposed system consists of four core modules: (1) Procurement Intent Recognition using contextualized transformer models; (2) joint entity and relation extraction for structured semantic understanding; (3) knowledge graph construction and clustering are the terms where entities and relations that have been extracted are validated, disambiguated, and checked for semantic compatibility before they are applied to the generations of the documents that are downstream.; and (4) a hybrid document generation strategy combining slot-based templates and large language model (LLM) rewriting. A high-quality Chinese procurement dataset containing over 8,000 annotated requirements and 5,000 formal tender documents was constructed to support model training and evaluation. Experimental results demonstrate that the proposed system achieves high accuracy in named entity recognition (F1 = 86.8%), intent classification (F1 = 90.6%), and triplet extraction (F1 = 84.5%), while generating fluent, professional tender documents with a ROUGE-1 score of 73.2% and BLEU-4 score of 61.7%.</p>