Research Interests: Reinforcement Learning, Computational Physics, Large Language Models, Human Computer Interaction
At Amazon, specifically AWS Kinesis and AWS AI Labs, I worked as a hands-on research engineer extending and operationalizing classic and machine learning algorithms.
Robust Random Cut Forest: is an algorithm invented at Amazon by scientists Sudipto Guha and Nina Mishra. I worked alongside them to extend use cases for this algorithm to work on graphs, for hotspot detection, for semi-supervised learning, and to provide greater explainability for anomaly detection. We deployed this as part of a new set of features for Kinesis Analytics
Sagemaker Ground Truth I was a founding engineer for Sagemaker Ground Truth and worked as a research engineer alongside Pietro Perona and Fedor Zhadanov’s team with a particular focus on extending support for the object detection modality. I also took a seminar in deep learning taught by Krzysztof Chalupka.
While spending a little time in Berlin, I met Meredith L. Patterson CEO of Special Circumstances, a group of hackers, security and machine learning researchers that participated on a range of DARPA contracts.
I participated as an applied scientist on AIMEE, was the (later stage) principal investigator for REMATH, and principal investigator for HARDEN.
Weird Mario: Intrepid Super Mario World players have found mechanisms for injecting arbitrary code by exploiting known bugs in the game to trigger execution of arbitrary code programmed with the use of only the game controller . Reproducing these behaviors with a reinforcement learning solution required implementing an agent that could reason with a sparse and deceptive reward.
REFUSR: I implemented this paper to recover OpenPLC programs from a small set of observed inputs and outputs. This project made me a lot more interested in using reinforcement learning for search.
PARX: I built a new under-approximate method of concolic execution aimed at uncovering sequences of inputs that resulted in violations of a model of a computer program specified by an end user.
During this time I formed my own company and sub-contracted to Special Circumstances. In addition to the work above we worked on
Bright Beams: We developed virtual diagnostic evaluated on the linear accelerator at LBNL.
Compositional Reasoning for LLMs: I provided support on a paper that outlined performing compositional reasoning using external knowledge during LLM inference.
Motion Stabilization for Quantum Gravimetry: I am currently working on a solution for increasing the sensitivity of a mobile absolute gravimeter.
FPGA Synthesis I am starting a commercial venture that aims to make FPGA programming ridiculously easy, bettering the odds of the hardware lottery faced by software engineers and researchers
I like to read a lot. My favorite periods in literature are 1970s Italian feminism, post-war Japanese literature, Latin boom and the global “I-novel” . I love Ozu, Cassavetes, and technicolor . Visual fan of abstract expressionism and brutalism.
I wish I was doing more work in game theory, self-supervision in reinforcement learning, and world model building in LLMs.
My intellectual hero is Ludwig Wittgenstein.