Hello, and welcome to the Hopline blog.
We’re a studio for deep tech and specialised systems: compilers and optimised backends, GPU and HPC workloads, formal verification, agentic control systems, and ML that has to run inside an air-gapped network. The common thread is that adding engineers doesn’t help. If correctness, performance or physics has already ruled out the obvious approach, a bigger team arrives at the same wall faster.
This space is where we write about that work.
What you can expect here
Not growth-hack listicles. The posts here will be the practical kind of thing we’d want to read ourselves: what actually happens when a multi-node job hits GPUs that don’t share an instruction set, how you measure something as slippery as plasticity loss, why probabilistic retrieval is the wrong tool when a false negative is legal exposure. Some of it will be about the tooling we publish, since we can talk about that freely.
What you won’t find is client work we haven’t been cleared to describe. Naming and publication get agreed at the start of an engagement, which means some of the most interesting things we do will show up here late, or in anonymised form, or never.
Why we write at all
Two reasons, both selfish in the good way.
First, writing forces clarity. If we can’t explain a decision simply, we probably don’t understand it well enough yet. Publishing keeps us honest.
Second, the things we figure out shouldn’t have to be re-figured-out by the next team. The same instinct that makes us publish the reusable parts of our tooling makes us want to write the reusable parts of our thinking down.
What’s next
Current engagements are described on the references page, the general-purpose tooling lands on GitHub, and what we learn lands here.
If you’ve got a problem with no obvious approach, book a scoping call. We reply within a day, and it’s one of the founders reading it.