AI-Powered Content Moderation and Harmful Content Detection in Streaming Media

Online video platforms like YouTube are flooded with new uploads every minute, which makes it hard to control harmful or sensitive content such as violence, nudity, or abusive speech. To tackle this issue, this study introduces SafeStream.ai, a simple web tool that helps viewers watch safer versions of videos. The system automatically spots and skips inappropriate parts using a mix of computer vision, text analysis, and emotion detection. It’s built with three main parts: a Next.js interface for users, a Flask service that runs the AI model, and a Neon Postgres database to save results. When a user adds a YouTube link, the system breaks the video into frames, checks each one for unsafe visuals via computer vision, and records time ranges where such content appears. Tests so far show about 67% accuracy in identifying harmful material. SafeStream.ai aims to give users control over what they see online while keeping the experience smooth and accessible. Future improvements will focus on increasing accuracy, speeding up processing, and training with more diverse data. The project also plans to go open-source to invite community input and make the internet a safer place for everyone.