July 13, 2026
SQL just got a main character arc
Show HN: I implemented a neural network in SQL
A coder taught a database to do AI, and the comments swung from eye-rolls to applause
TLDR: A developer showed off a way to make a database perform AI-style learning math, which is unusual because databases normally just store and fetch information. Commenters went from mocking the idea to praising it, with jokes, confusion, and meme comparisons stealing the show.
A Hacker News post about someone building a neural network inside SQL — the language usually used to pull data from databases — sent the comments straight into "this is cursed" meets "wait... this is actually brilliant" territory. For non-tech folks: imagine using a spreadsheet-style data tool to do the kind of pattern-spotting math usually handled by specialized AI software. That alone was enough to make people do a double take.
The biggest mood swing came from readers who started out skeptical and ended up weirdly impressed. One commenter admitted they "rolled my eyes" at the idea, only to come away convinced after digging into the code. Another took the more relatable route, joking that this was their sign to retreat back to making boring everyday app features because they only understood "about 10%" of what was happening. Honestly? Extremely real.
And then came the jokes. One of the funniest hot takes declared this was "more reliable than implementing SQL in neural networks," which instantly triggered the perfect internet response from another user: "Why? lol" That tiny exchange delivered the thread’s mini-drama — equal parts smart-aleck flex and confused heckling. Another commenter compared the whole thing to those goofy "can it run Doom?" stunts, except this time the challenge was "can it run AI in a database?" Bottom line: the project impressed people, but the comments turned it into a full-on spectacle of disbelief, nerdy admiration, and meme energy.
Key Points
- •The article is a Show HN post about implementing a neural network in SQL.
- •The implementation represents each neural-network layer as a table chunk.
- •Weights are stored using input and output indices, noted as `(inp_i, out_i)`.
- •The layer representation also includes a bias term.
- •The provided article content focuses on the structural representation of layers in SQL tables.