For the past few months our team has been working on an agentic framework that takes large, expensive SQL workloads and compiles them to Rust, backed by a formal proof of correctness. This is the sort of work I used to do by hand as a systems performance engineer, but now it happens in hours instead of months.

We first redid our internal pipelines with this technique and they're now 10x faster and 3x cheaper. Next, we went live with our Snowflake and Databricks SQL customers. They're seeing similar results with critical production workloads that originally cost 6-7 figures per year.

We'll be publishing benchmarks and code samples over the coming weeks; for now, you can read a bit more (and watch our anime launch trailer) on the site.

Happy to answer questions about how it all works.