Customizable Mini LLM Builder

A platform that lets anyone train a small language model (9M parameters) with their own data in minutes on free compute. Users can swap in custom personalities, training data, or use cases without needing ML expertise. Target users: educators, hobbyists, indie developers, and anyone curious about how LLMs work.

The explosion of interest in local and small-footprint AI models — driven by Ollama, LM Studio, and the Llama/Phi model families — has created a genuine audience of developers who want to understand and customize models without GPU bills. Hugging Face already offers fine-tuning workflows and free compute via Spaces and AutoTrain, which covers a lot of this ground for technical users; the differentiation here would have to live entirely in UX simplicity for non-technical audiences. A $500–2k/mo revenue band is plausible only if a freemium-to-paid conversion funnel works on hobbyists and educators, which is notoriously hard — these segments explore enthusiastically and pay reluctantly. The biggest risk is that 9M-parameter models produce outputs too weak to feel useful, leaving users underwhelmed and churning before they've seen enough value to pay anything.