I watched one Formula 1 video and suddenly YouTube decided my entire personality is “Max Verstappen fanclub.” And because I’m a developer (and this is how we process emotions), I built my own recommendation algorithm. The goal wasn’t “more watch time” or “more dopamine”—I wanted a system that recommends videos based on what’s actually going to inspire me. No random F1 interviews, no Andrew Tate detours, and ideally fewer viral things I don’t care about. In this build, I pulled videos from the YouTube API using Python, stored everything in a SQL-like database (because yes, I overengineered a minor inconvenience), and trained a random forest model on my own yes/no ratings plus feedback sentiment, keywords, duration, and view counts. Then I used ChatLLM’s Deep Agent to analyze my raw rating data and it exposed some brutal truths—like how I’m apparently anti-tutorial and weirdly biased toward niche, low-view content. I rebuilt my search queries around “creative coding experiments” and “AI project showcases,” shipped a UI that looks suspiciously like YouTube, and finally tested my algorithm against YouTube’s homepage. Spoiler: MyTube won… which is both satisfying and slightly unhinged.