I go by nikete in everything but the most formal of documents.

Broadly, I am motivated to work towards the improvement of the understanding. By desining, building and understanding the computational engines that mediate collective cognition. Methodologically my tools come mostly from economics and machine learning. I enjoy theory driven work and oportunistic experimentation. I run a small consultancy, Amurado Research OU, focused on commercial applications of machine learning and economics. I am also a research affiliate of the Laboratory for Computational Physiology at MIT, where I try to design better algorithms and mechanisms for finding which treatments to use in the Intensive Care Unit.

My current applied focus is on medicine. How can we design better ways of evaluating proposed policies given the existing observational data? I believe that the current way Reinforcement Learning algorithms in medicine are evaluated does not lead to a collective acumulation of knowledge about what effective policies are. I am trying to find better ways. While the ICU is my current “model institution”, but I have also studied various markets (betting, cryptocurrencies, internet advertising) and non-market institutions (various comunes and collectives). I would love to do some experiments about collecticve cognition around nutrition.

I wrote a PhD thesis, on algoirthms and mechanisms for aiding decision making designed to preserve subjects freedom of choice. There are two main results in it: 1) when you are learning from the rewards received in a bandit setting and the action that is taken is not the one the algorithm chose, by paying attention to both there are situations where you can do arbitrarily better with only a multiplicative worst case cost. 2) When you are getting advice to inform a decision that you want to mantain ultimate control over, allowing the advice givers to self select so one of them receives all information and “owns” the advice can do much better than trying to use “prediction market” mechanisms directly.

Google scholar has a good first approximation to my academic output.

Contact me