The use of AI-driven agricultural innovations in India to improve social enterprises is highlighted here in the research paper with a focus on the identification and classification of crops and optimal irrigation. The study uses various machine learning algorithms including Convolutional Neural Networks, RNN, Decision Trees, K-Means Clustering, and GBM among others to identify the best performing algorithms over various iterations. The average accuracy across all CNN models was 92.4%, with precision and recall values between 90.5% and 93.5%. In the case of RNN models, the Mean Absolute Percentage Error was 3.4% and Root Mean Squared Error. For Decision Trees, Gini Impurity and Information Gain values were similarly high, suggesting effective classification performance. The K-Means Clustering method resulted in an average Silhouette Score of 0.61 and an average Inertia of 350 in Crop Irrigation Optimization. Low scores were also demonstrated by the GBM model, with Mean Squared Error averaging 0.12, and Log Loss score averaging 0.32. These results indicate that AI-based innovations have the potential to enhance agricultural productivity, sustainability, and resilience in India, helping smallholders and social enterprises.