This new research advances the NLG technique in the automation of financial report generation using methods that are deep learning architectures combined with advanced natural language generation. A new hybrid system which is one step ahead of the prevailing template-based techniques provides structured templates and intelligent models of language thus producing comprehensive, accurate, and contextually relevant financial reports, overcoming deficiencies in currently found techniques for automated report generation while being in line with the requirements of reporting standards. A custom transformer-based architecture, to which financial domain-specific embeddings were added, was proposed. This ensured 98% accuracy in numerical interpretation and helped maintain high linguistic coherence. Experimental results illustrating large improvements over traditional approaches were presented: these cut the time for report generation by 85% yet still maintained standards of quality. Our system was able to process more than 1,000 reports across multiple sectors with robust performance in the processing of complex financial narratives and numerical analyses. Achieving a BLEU score of 0.85 and a ROUGE score of 0.82 against other current state-of-the-art approaches, this work extends the state of the art by developing an efficient and dependable method for the automation of financial reporting, without losing sight of the conditions of the industry and achieving the maximum reduction of human intervention and all associated errors. Different datasets were used to train the model including: yfinance to fetch financial data from Yahoo Finance, financial phrase bank and dataset from Kaggle website