This research introduces a novel way for identifying offline handwritten Gujarati conjunct consonants. This paper contains 960 distinct classes of handwritten Gujarati conjunct consonants. The process of deep learning is the technique utilized for identity identification based on artificial neural networks. Deep learning requires a precise emphasis on key visual attributes for self-recognition, with minimal reliance on the software. Deep learning enhances the precision and speed of recognizing joint Gujarati handwritten characters. EfficientNet, a type of convolutional neural network (CNN), is utilized along with pre-processing and normalization techniques to identify handwritten Gujarati conjunct letters. We comprehensively analyze the performance of several CNN architectures. Segmentation is done before recognition of handwritten Gujarati conjunct characters. Conjunct characters are divided into testing and training components prior to recognition. We implemented the proposed model to test and assess the system's performance. We utilized OpenCV, TensorFlow, and Keras libraries in Python to obtain superior performance compared to current methods. Utilize augmentation to develop and generate over 800,000 example photos. Utilizing the EfficientNet-based model can obtain an accuracy of over 83.88% with the proposed model.