Show HN: I trained a small local model to translate natural language to CLI (github.com)

A new tool, Zest CLI, has been unveiled to translate natural language into shell commands using small local language models, specifically designed for macOS users. This innovative application runs entirely offline on CPU, ensuring maximum privacy by eliminating reliance on cloud services, API keys, and telemetry—meaning users' commands and queries remain securely on their devices. The system utilizes purpose-trained small language models (SLMs) optimized for command-line interpretation, making it accessible without requiring specialized hardware or internet connectivity. The significance of Zest CLI lies in its commitment to user privacy and ease of use, catering to those who may be concerned about data privacy while still needing efficient command-line interaction. The training pipeline does use Google Cloud Storage for dataset management and fine-tuning within Google Colab but keeps this process separate from the actual usage of the tool, wherein all operations occur locally. Utilizing a Medallion Architecture pattern for data storage ensures high data quality and reproducibility, with structured layers for raw data processing and final training datasets, demonstrating a robust approach to responsible AI development.