Alessandro Duico
Hey, Alessandro here!
I'm passionate about RL, LLM post-training, and computer vision.
I currently work on text-to-speech at smallest.ai.
Before that I co-founded elephantroom.ai.
I graduated from TU Delft with a thesis on predicting the most replayed moments in videos, arXiv:2309.06102, advised by Jan van Gemert and Ombretta Strafforello.
Recent writing
- Deriving PPO from First Principles
An interactive walkthrough: building the PPO objective piece by piece from scratch.