Show HN: what-time Neural Parser for Schedule extraction in English, Hindi (github.com)

A new open-source tool, "what-time," has been introduced for neural schedule extraction in English, Hindi, and Hinglish. This on-device software allows users to convert natural language phrases such as "call mom on Sunday evening" or "कल शाम को आठ बजे" into accurate dates, time ranges, and recurrence rules without requiring server access. With a transformer model consisting of around 144,000 parameters, it ensures precise tagging of tokens and builds schedules through a deterministic compiler that integrates a timezone database. Notably, the tool is designed to produce detailed diagnostics for any parsing failures, rather than silently returning incorrect data. This development is significant for the AI/ML community as it showcases advancements in natural language processing, particularly for multilingual applications. By functioning offline, "what-time" prioritizes user privacy and enhances user experience on various devices. The tool is versatile, offering a Rust library, CLI, and an easy-to-use playground, while supporting bulk processing with GPU acceleration for performance efficiency. Its approach to training and error handling sets a benchmark for future models, emphasizing reliable and interpretable outcomes in NLP tasks, especially in applications that require precise temporal understanding.