LiminalML

Six-stage study sessions for mastering ML & SWE

Education

Developer Tools

Artificial Intelligence

Hey Product Hunt 👋 I'm Oliver, an ML engineer at a stealth startup. I built LiminalML for myself first. The pattern I kept hitting was when I wanted to deeply understand things like transformers, MTL approaches for my research, or anything really when I was studying for interviews, I'd constantly have so many different resources that explained topics I liked cleanly, and I was tired of everything being in different places like Stanford Online lectures, my university notes from lectures, or just the research papers themselves. Overtime I figured out a structure that fit my way of learning really well that I wished every educational resource had. That's where I landed on the concept of six stages to learn any topic I wanted to really deeply, to the level of mastery. This helped me retain a lot of information for interviews, as in ML specifically people are expected to know everything from classical ML to CS fundamentals to post training technique to attention mechanisms and beyond, so for a new grad, it was a lot, especially while finishing up university courses. So LiminalML puts every topic through the same six stages: the big-picture intuition, visualization, the math derived step by step, code you can actually run and edit, the questions an interviewer would push on, and a retrieval drill at the end to check it stuck. Math, diagrams, and code render inline, and every session saves so you can come back to a thread later. It's built on Claude under the hood. It's for anyone going deep on ML or SWE. Interview prep is the sharpest use case as you can upload your resume for context across sessions and have star stories generated based off of it and that tailors the questions asked during the retrieval stage. I still use it often to rebuild fundamentals I've gotten rusty on. If you want to see the format before signing up, the concept pages are free to read with no account. It's just me building this, and it's early, so I'd genuinely value feedback on where the explanations should go deeper, or where the six-stage structure gets in the way. Happy to get into how it's built, too.