Tiny LLM Training Platform

A platform that lets non-ML engineers train small language models (9-50M parameters) to understand how transformers work. Users can customize synthetic training data, swap personality traits, and train models in minutes on free compute. Target: students, curious developers, and AI enthusiasts.

The surge in AI literacy demand is real — bootcamps, university programs, and self-taught developers are all scrambling to build intuition for how transformers actually work, and hands-on training beats reading papers. Hugging Face covers this space partially with their AutoTrain and course materials, but it's oriented toward practitioners who already know what they're doing, not true beginners. The $1k–5k/mo revenue band is realistic only if free compute costs stay contained, which is the central tension — subsidizing GPU time for curious hobbyists who churn quickly is a brutal unit economics problem, and that's also the biggest risk: the moment free compute gets expensive or rate-limited, the core value proposition collapses and there's not enough willingness-to-pay among students to cover it.