Show HN: Json2vec – build/train/deploy models with nested data structures

Category: library

Tags: pytorch, deep-learning, json-schema

Score: 6.3/10 (Innovation: 6, Technical: 6, Documentation: 8, Utility: 5)

Json2vec is a Python library that builds PyTorch/Lightning models directly from JSON-like schemas, enabling predictive modeling on nested data structures without flattening. Its core innovation is using the schema as both a data contract and model architecture blueprint, with built-in support for supervised, masked, and unsupervised learning modes. The project is interesting for its approach to handling hierarchical relationships and mixed datatypes in business data.

Target audience: data scientists, machine learning engineers, backend devs

Repository: https://github.com/json2vec/json2vec · Python · Apache-2.0 · 8 stars

View on Hacker News