This paper introduces a hybrid CNN-BLSTM model designed for recognizing handwritten Devanagari characters, emphasizing transparency and reproducibility. By integrating convolutional neural networks (CNNs) and bidirectional long short-term memory (BLSTM) networks, the model achieves robustness in character recognition. The CNN component extracts hierarchical representations of input images, capturing essential patterns and structures, while the BLSTM component processes sequences bidirectionally, understanding dependencies and relationships between characters. The model architecture undergoes structured data preprocessing, including grayscale conversion, resizing, and normalization, ensuring standardized inputs for effective training and evaluation. The CNN architecture includes two convolutional layers with ReLU activation and max pooling, followed by a flattening layer. The BLSTM layer processes flattened features bidirectionally to capture sequential dependencies. Key components of BLSTM include forget gate, input gate, candidate cell state, update cell state, output gate, and hidden state. The Time-Distributed Dense layer maps BLSTM outputs to class probabilities, aiding character recognition. Accuracy achieved is 95.29–99.55% on different datasets. Overall, the model effectively learns and recognizes Devanagari characters by leveraging both CNN and BLSTM architectures.