Enhancing Data Entry Efficiency by Combining Generative AI and Handwritten Digit Recognition
Ren Wu, Kurumi Tsuruta · 2025
This paper proposes a method to improve data entry efficiency in application development by combining Generative AI with a handwritten digit recognition architecture. Our approach leverages the generative capabilities of GPT-4o along with the digit recognition proficiency of a LeNet-5 model trained on the MNIST dataset. While individual evaluations on a 700-digit dataset yielded accuracies of 85.14% and 96.45% for gpt-4o and LeNet-5, respectively, the combined system achieved an accuracy of 82.71% with a high confidence level of 99.86%. Additionally, it has the potential to eliminate the need for manual data entry for up to 80% of digit instances (20 out of 25 cases).