Johann Ross-Jimenez

At Ousia Research, I fine-tune LLMs and publish datasets, benchmarks, and agent tooling as open-source work. I’ve released four fine-tuned models and 12 public datasets through the Hugging Face Hub.

My Aureth releases and community derivatives have passed 4,700 cumulative downloads. I also built the training datasets end to end, from collection and cleaning through formatting and Hub documentation.

I was first author of a 25-page paper and benchmark studying narrative identity effects in LLM reasoning. The work tested five models across 45 conditions and 2,608 trials, and derived invertible dose-response steering equations.

I’m a credited contributor to the Hermes Agent ecosystem at NousResearch, where a plugin based on my work was accepted into the official catalog. I also ported the Muse Gadgets SDK to an ESP32-S3 touch LCD board, including its driver, avatar renderer, and gesture engine.