About Me
I'm a computer scientist with a background that cuts across low-latency systems, computational biology, and large-scale deep learning. The common thread is building things that have to work correctly under hard constraints — whether that's a matching engine processing millions of orders per second, a distributed simulation of emergent biological behavior, or a training pipeline for a billion-parameter model.
For a large part of my career I built trading infrastructure in C++ — matching engines, order management systems, market data pipelines and high frequency trading robots.
My PhD at Tufts took me in a different direction. I built a full agent-based modeling framework to study how complex behaviors emerge in different scales of biological systems: from cells to organisms to animal populations. The questions were about emergence, collective intelligence, and how global order arises without central coordination — questions I find myself returning to now when thinking about neural networks.
For the past years I've been working on large-scale deep learning, training and evaluating foundation models for enterprise applications. This means the full stack: dataset curation, architecture experimentation, parallelism strategies, training efficiency, evaluation frameworks, and production deployment. I care especially about what happens inside these models — how they organize computation, what structures emerge from training, and what it would mean for them to genuinely represent the world.
That last question pulls me toward interpretability and toward philosophy. I'm currently writing about intentionality in large language models — whether something like aboutness can be attributed to learned representations, and what the right conceptual framework for thinking about this looks like. I draw on work in mechanistic interpretability, philosophy of mind (Brentano, Dennett, Chalmers), and ideas from complex systems and artificial life.
You can reach me at giordanobsf (at) gmail (dot) com or find me on [LinkedIn / Twitter / GitHub].