Background Follicular lymphoma is a common lymphoma of germinal center B cells. Accurate grading of follicular lymphoma is challenging, as it relies on accurate identification and quantitation of neoplastic centrocytes and centroblasts. within the tissue sections. Grading is consequential in determining plans and treatments for patients, however there is a high degree of inter-observer variability among practicing physicians. Artificial intelligence (AI) approaches present an opportunity to refine the grading process for follicular lymphoma and provide better reproducibility compared to that achieved by pathologists and ease the workload of physicians in a reviewing a high volume of slides. The aim of this study is to develop and incorporate an AI-driven follicular lymphoma grading model into routine pathology practice to improve patient care. Methods 1313 H&E-stained Whole Slide Images (WSI) with diagnoses of follicular lymphoma from 441 distinct patients submitted to hematopathology service in our institution during 2019- 2021 were collected. The WSI were categorized into grades 1-2, 3A, and 3B based on clinical pathology report and then split into training (70%), validation (15%), and test (15%) datasets at individual patient level following by pre-processing. The WSIs were processed at resolutions of 20x, 10x, and 5x. A pre-trained ResNet-50 model was applied to the training set to extract features at the tile level. Multiple Instance Learning (MIL) with a Random Forest (RF) classifier was utilized to determine the slide's overall grade. Results The neural network model classified the WSI into grades 1-2, 3A, or 3B with 85% accuracy overall. The tile-level ROC curves displayed AUROC scores of 0.95 for both 20x resolutions and 0.96 for the 5x resolution. Precisions were 87% for 20x, 84% for 10x, and 86% for 5x resolutions. For slide-level classification, we used concatenated features derived from the tile level. The slide-level AUROC revealed an AUC score of 0.94, accompanied by a precision of 80%. Conclusion Our model has demonstrated efficacy and precision in predicting the grades of follicular lymphoma. Our model's accuracy surpasses reported inter-reader consistency among practicing pathologists. This model offers promise as an initial grading tool, facilitating pathologists' tasks by enabling them to validate preliminary findings. With use of this tool, we can ease the of grading and review of the high volume of slides that pathologists analyze daily.