A Hacker News user posted a Show HN demonstration of a neural network implemented entirely in SQL.

The post, titled “Show HN: I implemented a neural network in SQL,” was shared on May 5, 2025.

The developer described an architecture that defines input preprocessing, hidden layer calculations, and output activation using standard SQL functions, all executed within a PostgreSQL database without external libraries.

According to the post, the model was trained on the MNIST digit dataset and achieved approximately 90 % accuracy on the test set.

Comments on the thread highlighted the approach's applicability and noted performance constraints and the absence of modern deep‑learning frameworks.

The author invites further experimentation and potential optimizations, leaving the community to evaluate the practicality of SQL‑based neural networks.