Ray Tracer in SQL
github.comIn the Age of AI coding, this lovely cool demos are no longer interesting anymore. I use to appreciate them by the craftsmanship and hacking aptitude they required, but now AI took away that joy of watch this for me.
Yeah. Remember when that guy made Doom in TypeScript types? That was incredible. This feels shallow and dull. Interesting idea and cool that ClickHouse -- a primarily human made piece of software -- can even do this but I agree it no longer does it for me.
Do you really think AI can make Doom in TypeScript types?
Yes: because it will simply steal (rather, has stolen) this guy's genuine, human, work.
Absolutely. These things are inherently pointless - what makes them cool is the human ingenuity required to achieve them. Remove that and it’s totally uninteresting.
That feeling has nothing to do with AI. That's how art has always been. Most of these were bad at being art even before AI.
A good reference for exactly what I mean would be the demoscene (both back then and now). You can watch a thousand of those and be totally underwhelmed, but every now and then you get one that totally blows your mind.
There's nothing wrong with seeking novelty, but there is something wrong being jaded about it.
Who are you replying to?
Coding with AI is not much different from coding without AI, just faster and with slightly less typing. But you still have to organize the work and split it into small items. If you ask a model to implement a ray tracer in SQL, it can do something one-shot, but it won't be as satisfactory as with an engineer.
Indeed, I am just waiting for the last drop, when AI tooling becomes good enough for handling machine code directly.
Note that a lot of programming can already be done via orchestration flows calling into mcp tools (replacing classical microservices).
All the no code/low code tools have migrated into it as their evolution.
Very cool! I did a similar (but much simpler!) experiment by implementing perspective projection via SQL, storing meshes (vertices, edges, faces), the camera position and the screen size in tables and building a single query that generates the SVG paths (including backface culling). Running via WASM SQlite inside the web browser. [1]
SELECT project(...) as x, project(...) as y
FROM model, vertex, camera, transform
WHERE clockwise AND clipped IN BETWEEN -1 AND 1
Very cool project. Next they should create a database with SQL. Querying my database's database with my database's SQL's SQL would be trippy.
SQL in SQL has never occurred to me.
Nice, but why ?
It is needed to find bottlenecks in the engine and optimize customer queries. For example, it gives things like: https://github.com/ClickHouse/ClickHouse/pull/108757
You can read more about this approach in my presentation: https://presentations.clickhouse.com/2026-openhouse-sf/great... and video: https://www.youtube.com/watch?v=FmS7VopaqNg
sir... this is hacker news
Good stress test for your SQL parser I guess.
To know that it can be done.
Why not?
I love all the "But Why?!?" and "It's LLM so it doesn't count" comments.
It is needed to find bottlenecks in the engine and optimize customer queries. For example, it gives things like: https://github.com/ClickHouse/ClickHouse/pull/108757
You can read more about this approach in my presentation: https://presentations.clickhouse.com/2026-openhouse-sf/great... and video: https://www.youtube.com/watch?v=FmS7VopaqNg
How do I join this awesome project?
Look, I love a good hack, but just. No. Why would you ever want to do this. The intersection of skill and masochism needed for this shouldn't exist and yet it does