Building Undermind.ai (2023).
We use LLMs to automate reasoning and accelerate scientific research.
Gave a talk at MIT physics describing the roadmap of Undermind.
Independent study (2022-2023). See Github/tomhartke.
Built an external memory module for a language model in python. The program extracts a structured knowledge graph from a set of known facts using GPT-3. The long term goal is to build an autonomous research assistant, which asks questions to fill gaps in the knowledge graph.
Brainstormed methods to shape the way information is stored in neural networks during training.
Implemented the AlphaGo zero algorithm using tree search and deep neural networks.
Quantum Simulation with Ultracold Atoms
Created a quantum gas of attractive fermion pairs under a microscope, a system which sheds light on high-temperature superconductivity.
Invented, patented, and demonstrated a new method of storing and manipulating quantum information in the vibrational motion of pairs of fermonic atoms trapped in an optical lattice.
Invented a novel method of quantum gas microscopy to probe a repulsive Fermi gas. Observed local doublon-hole pairs which are a signature of the super-exchange mechanism that drives magnetic order of electrons in two-dimensional materials.
Measured the spin transport properties of a strongly-correlated Mott insulator of ultracold fermions. Such behavior is difficult to probe in solid state materials.
Quantum Control of Single Electrons
Demonstrated coupling of the quantized motion of a single electron, vibrational phonons in a nanowire, and photons in a superconducting cavity by building a cavity-coupled double-quantum-dot device (undergraduate thesis in 2017).
Upgraded a reactor in an attempt to grow nanowires with superconducting shells (summer 2015).
Built simple models for quantum limits to measurement damage with Ed Boyden at MIT.
Simulated space plasma dynamics with Marc Swisdak at University of Maryland (summer 2016).
Explored chaotic classical and quantum systems with Herman Verlinde at Princeton (spring 2016).
Measured the hypoxic response of stem cells at Celgene Cellular Therapeutics (summer 2012).
Studied the deterioration of ceramic high-temperature thermal barrier coatings in turbine engines using scanning electron microscopy at Princeton (summer 2011).
Biotech, neuroscience, and other research with Xapiens@MIT (2019-2022).
Physics education outreach as a board member of the nonprofit Physics Unlimited (2018-2022), which also runs an annual competition for high school students. I wrote the 2016 competition problems on entropy and statistical mechanics.
Robotics in high school. I co-designed the 2014 New Jersey Panasonic Creative Design Challenge.
Economics, history, sci-fi novels, archaeology, and ancient civilizations.