About Pico

Pico was founded at the University of Cambridge by Richard Diehl Martinez and Paula Buttery with a simple conviction: training language models should be a science, not an art.

Built around two core libraries — pico-train for model training and pico-analyze for in-depth analysis — Pico creates a sandbox for researchers to develop and test new ideas in efficient training, interpretability, and model behaviour.

The project is currently led by Suchir Salhan, who drives the research direction and day-to-day development. This journal is where we share our findings, technical deep-dives, and lessons learned from building and studying small-scale language models.

Team

Richard Diehl Martinez

Co-Founder

Richard co-founded Pico at the University of Cambridge, where he led the development of modular tooling for systematic language model research. His work focuses on understanding training dynamics and building efficient, reproducible ML infrastructure.

Paula Buttery

Co-Founder

Paula is a Professor at the University of Cambridge Department of Computer Science and Technology. Her research spans computational linguistics and NLP, and she co-founded Pico to bridge the gap between rigorous linguistic research and modern language model development.

Suchir Salhan

Project Lead

Suchir is a Computer Science PhD candidate at the University of Cambridge, working on data-efficient small language models. He currently leads day-to-day development and research direction across the Pico project.