I'm going through CMU 15-721 and building wsql[1]; a GPU-accelerated query engine. It uses wgpu for cross-platform portability and implements modular pattern, heavily inspired from Sirius[2] paper.
[1]: https://github.com/fastingrat/wsql/
[2]: https://vldb.org/cidrdb/papers/2026/p12-yogatama.pdf