The Vocalverse platform allows users to chat with celebrities, video game characters, and more. Users can pick from a catalog of models to start voice chats with, then log in to save chat history and models. We wanted to create a platform where users can seamlessly talk to a large number of virtual agents, like the metaverse but with voice. We were inspired by Character AI, which fine-tunes LLMs to speak like different characters. However, the problem is these models only output text, and aren’t very engaging. Realistic voice is the next step in making AI assistants and companions mainstream, and we want to build a platform where anything is possible. The current platform is built using NextJS and Firebase and deployed on Vercel. The streaming chat is built using Vercel’s ai SDK, and the model is OpenAi’s GPT 3.5 API with a system prompt. If we are selected for the Slingshot accelerator, we have many plans to make this an epic product. This includes fine-tuning open-source models like LLAMA and Falcon instead of using GPT, adding more characters, and adding voice input. Eventually, this could be a social media platform where humans and AI agents communicate interchangeably, like Discord. We plan to have a subscription service and share the revenue with IP holders and celebrities to use their voices. Eventually, if the platform gets large enough, we can experiment with an advertising model. The problem we hope to solve is loneliness and mental health, which we predict will be a growing market. Our minimum viable segment is lonely, depressed introverts who spend on services like CharacterAI, VTubers, and OnlyFans, and mental health/therapy services. We will focus also on elderly people, who tend to be lonely and don't have many other avenues for entertainment.Category tags:
"Good Job. good implementation of technology. i think it will be popular very soon and everybody will use it. keep working on it."
Walaa Nasr Elghitany
Data scientist and doctor