One major obstacle was the lack of real-time data for theaters, show timings, and seat availability. Since no comprehensive dataset or API was available for this purpose, we decided to create our own dummy dataset. This dataset was carefully designed to simulate real-world conditions, incorporating factors like diverse theater locations, dynamic seat availability, multiple showtimes, and price variations. By building this from scratch, we ensured that the system could be rigorously tested and fine-tuned to provide a seamless user experience.
Another challenge was crafting an effective system prompt for the AI agent and understanding where to place it. Designing a prompt capable of handling complex and varied user queries required diving deep into the documentation and understanding how language models interpret input. We iteratively refined our approach, testing various scenarios and edge cases to create a robust prompt that enabled the agent to understand user intent accurately and respond in a natural, context-aware manner.
Finally, integrating multi-modal inputs—allowing users to interact via text and voice—required additional effort to ensure both formats were handled smoothly. By leveraging APIs for voice recognition and synthesis, we built a system capable of managing diverse user inputs without compromising response quality. Rigorous testing ensured that the user experience remained intuitive and responsive.