Who am I?

My name is Nymph. I'm an infrastructure and backend engineer working on performance engineering and distributed systems, with a keen interest in mechanistic interpretability. Currently my main project is building steering vector support for vLLM in an effort to bridge the gap between research techniques and production code. Research won't ever get applied in production unless someone takes the time to make it fast and reliable, and I want to be a part of that.

Professionally I work at Whetstone Research where I own the indexing and data infrastructure. I have 6 years of experience managing distributed databases, ETL pipelines, and backend APIs. For several years I have been the sole engineer on call to diagnose and fix problems in the prod data pipelines to maintain correctness and uptime.

Outside of work I spend most of my spare time contributing to open source AI infrastructure. In addition to the vLLM steering vector support, I am also working on a verifiers environment and prime-rl training setup for RL on code optimization using mechanistic reward signals from perf traces instead of wall-clock time.

I'm also interested in biotech and the ability of AI to accelerate science and medicine. I believe it is one of the most important avenues for technological advancement to unambiguously improve our lives. I have been writing a CLI for easily downloading and embedding data required to run PathoBench. I'm doing this work to help MedARC evaluate OpenMidnight and make it easier for others to benchmark their own models.