Timetable generation in academic institutions is a complex combinatorial problem that involves satisfying a multitude of hard and soft constraints related to courses, classrooms, time slots, and instructor availability. Traditional approaches often rely on heuristics or optimization methods that can become inefficient or infeasible as the problem size scales. In this paper, we present a deep learning-based generative methodology using Long Short-Term Memory (LSTM) networks to address the academic scheduling problem. By modeling timetable construction as a conditional sequence generation task, our approach learns temporal dependencies, resource usage patterns, and constraint structures directly from data. Each schedule entry is treated as a structured tuple encompassing the relevant scheduling attributes, and the LSTM model is trained to predict the next valid entry in an autoregressive manner. To ensure practical applicability, the output of the model is post-processed through a constraint-checking framework that enforces institutional policies such as batchwise lab assignments, room exclusivity, teacher non-overlap, and fixed lunch breaks. Extensive experimentation on synthetic datasets simulating real-world scenarios shows that our system consistently generates complete and conflict-free timetables with zero collision rate and sub-second generation times per section. The architecture is scalable, modular, and adaptable to new courses or departments without retraining. This generative modeling framework demonstrates strong potential for real-time deployment in large academic environments, offering both automation and intelligent scheduling capabilities.