Show HN: Compile English specs into 22 MB neural functions that run locally

Category: ai-ml

Tags: neural-compilation, local-ai, function-generation

Score: 7.0/10 (Innovation: 8, Technical: 7, Documentation: 5, Utility: 6)

PAW compiles English function descriptions into small, local neural programs (22 MB) that run offline without API keys. It's interesting because it uses LoRA adapters on a fixed pretrained model to create specialized functions, achieving better accuracy than direct prompting of larger models for specific tasks like classification and JSON repair.

Target audience: backend devs, ai engineers, devops

View on Hacker News