Joe Davison

Hi, I'm Joe. I'm an AI/ML engineer and researcher focused on LLM reasoning systems: reliable agents, inference scaling, and reinforcement learning with verifiable rewards.

I lead AI work at BambooHR (shipping the next generation of Ask BambooHR). Previously I was on the Hugging Face science team—you may know my work if you've ever encountered the zero-shot classification pipeline or the research guide on calulating fixed-length model perplexity. I also had a stint as a Senior ML Scientist at Enveda Biosciences, applying Transformers to mass-spec structure prediction for drug discovery.

I earned an MS in Data Science from Harvard, working with faculty including Sasha Rush and Finale Doshi-Velez, and spent a year in doctoral studies at the University of Utah with Vivek Srikumar.

Work that excites me right now includes RLVR research like ProRL from NVIDIA, the open sourcing of GPT-OSS weights and Harmony, and recent results on FrontierMath and IMO.
Twitter LinkedIn GitHub 🤗 Hub Scholar