In this paper, we have shown the development of an English–Odia machine translation (MT) system. For training, the system NLLBv1 dataset was taken. A baseline NMT was trained using this dataset. During experiments, it was identified that the system had some very complex sentences which produced improper results. It was also noted that the system was unable to problem-handle named entities. To address these issues, we developed a modified NMT system which had a subsystem. This subsystem simplified the complex English text into simple English text. This was an NMT system that was trained on the sentences taken from Wikipedia. Both NMT systems were evaluated using the standard MT evaluation metrics, viz BLEU, chrF and COMET. The evaluation was done on 600 sentences which were divided into 6 documents of 100 sentences each. It was found that our approach was able to produce better translations than the baseline NMT system. The system-level scores of both approaches were also taken. It was found that our approach was again able to produce better results. BLEU, chrF and COMET scores of baseline systems were 0.124983, 0.344197 and 0.471616, respectively, and that of our approach were 0.155812, 0.372303 and 0.634338, respectively.