I open-sourced the system behind my multilingual educational video generator

I’ve been working on an open-source project called SarasFlow. You give it a topic, and it moves through a series of steps to create a complete educational video: writing the script, generating the narration, creating the visuals, adding word-timed subtitles, and assembling the final video. I wanted each part of the system to be replaceable rather than locking everything behind one model or provider. There’s still a lot to improve around job orchestration, retries, provider selection, quality checks, and batch generation. I’d love feedback from people building AI systems with multiple coordinated stages. Contributions are welcome, and a star would help more developers find the repo.