I am Fede, a Senior ML Scientist at TogetherAI (formerly at OpenEvidence and Stanford). I am working on post-training in LLMs, mainly self-improving and self-evolving AI Agents.

I am interested in self-evolving agents and their application in science. I co-invented TextGrad (Nature 2025), a library for autograd with textual feedback, and TTT-Discover, achieving new scientific discoveries on open problems through test-time training. I also co-organized Agents4Science, the first AI-first scientific conference (featured in Science, Nature, MIT Technology Review, and Science News).

My safety research has been adopted by Meta, OpenAI and Anthropic, while FashionCLIP serves as a backend for numerous fashion recommendation systems, accumulating 100 million downloads.

My work has been published in top venues such as Nature, Nature Machine Intelligence, ICML, ICLR, and NAACL. Full publication list on Google Scholar.