Conversational AI has made significant advancements in recent years, enabling it to simulate human-like conversation and understand context. The technology is becoming increasingly sophisticated, allowing chatbots to provide more tailored and relevant answers to users' questions. However, the current training method of relying on vast numbers of documents downloaded from the web may not lead to natural conversations. To overcome this challenge, conversational AI needs to be trained on actual conversations, such as video calls, which contain valuable information including audio and visual cues that can provide input for training AI systems. Training on video calls could improve the ability of chatbots to understand nuances of human speech and recognize emotional states, leading to more effective communication in natural two-way conversations. A corpus of video calls can be used to train large language models and address the lack of training data identified as a challenge with current state of AI, enabling companies to create artificial intelligence agents for specific use cases such as customer service or telehealth.