Show HN: Train a 230KB text classifier from 50 examples – no API keys, no GPU (github.com)

Expressible AI has launched a groundbreaking tool that enables users to train a local text classifier with just 50 labeled examples, making complex machine learning tasks more accessible than ever. The tool, named "Distill," operates without relying on cloud services, API keys, or external computational resources like GPUs. Instead, it empowers users to keep their data confidential by processing classification tasks entirely on their local machines. The model, which is approximately 230KB in size, boasts an impressive accuracy rate of up to 95% for well-defined tasks, such as classifying contract clauses or support tickets. This innovative approach is particularly significant for industries with stringent data privacy requirements, including healthcare and legal sectors, as it allows for seamless classification while ensuring that sensitive information never leaves the user’s system. The ease of use—requiring no machine learning expertise and offering a straightforward CLI—facilitates quick setup and iterative improvement through a review and retraining process. By leveraging a tiny yet effective embedding model that encodes textual meaning locally, Distill stands to revolutionize how businesses handle and analyze text data while maintaining control and security.