This is still a draft, but I wanted to register the launch of kinderovo.com: a personal project where I worked through messy audio, transcription edge cases, and the practical limits of large language models while analyzing and structuring a podcast archive I genuinely enjoy. It became a good excuse to learn more about multimodal pipelines, how far current models can go when processing real-world audio, and where they still need help. I will write a much better and more detailed post about it soon.