Deep learning is an arm of Artificial Intelligence that uses deep neural networks to achieve artificial intelligence. It has made its mark in computer vision, speech recognition, language processing, and automatic engines. Google made a significant contribution to AI technologies by releasing TensorFlow (TF), its proprietary AI platform, in 2015, as an open-source software library to define, train and deploy learning models, including Machine Learning and Deep Learning. In this study, we aim to improve software estimation using the most recent deep learning paradigms. We employ TensorFlow and a high-level wrapper API to TF and evaluate a composite hyper-parameter tuning method employing the Cartesian grid and random search. We observe significant performance improvement, achieved (29.8%) from the base model, using the hybrid hyper-parameter tuning methodology. However, even While literature reports significant performance in cognitive imaging with TF and Keras, we have not been able to validate any substantial improvement in prediction, in the case of a software effort estimation data such as ISBSG 2018 by employing these techniques. © 2020 IJSTR.