The model that didn't exist, so you made it yourself
huggingface ·
This article details how the author utilized ML Intern, an AI agent within HuggingChat, to create six custom machine learning models in just a few days for approximately $103 in total compute costs. The process began with identifying specific needs, such as a lightweight prompt rewriter or specialized image generation capabilities, which were not readily available in existing public models. By providing detailed prompts that included verified facts, budget caps, and requirements for baseline testing, the author guided ML Intern to handle dataset preparation, training, evaluation, and publication on the Hugging Face Hub. Key projects included a citrus disease detection model, a LoRA for generating the Hugging Face mascot Huggy, and camera-angle editing tools for Qwen-Image. The author also demonstrated distillation techniques to reduce inference steps for browser-based models like Agate. This workflow highlights how users can now build tailored AI solutions efficiently without needing extensive manual engineering or large budgets.
huggingface