Hi, we're hiloop
Autoresearch as a service.
We turn a measurable target into thousands of directed experiments, and return a better model or algorithm. You bring the metric and a way to score it. We run the loop.
Backed byY Combinator
Trusted by
Hi, we're hiloop
We turn a measurable target into thousands of directed experiments, and return a better model or algorithm. You bring the metric and a way to score it. We run the loop.
What we've worked on
A model that has to score higher on an eval you own. We run the loop against that metric and return the weights.
Inference and kernel work where cost and throughput are the target: quantization, scheduling, speculative decoding.
The environments an agent trains against, generated and scored by the loop instead of hand-built one at a time.
Detection and decoding against a labelled corpus, where accuracy is measured the same way every run.
Mailing list
Progress notes from the founders — for investors, friends, and the curious. A few emails a year, one-click unsubscribe.