Developed a Pytorch-native framework that reduces implementation overhead by abstracting distinct components of reinforcement learning algorithms into orthogonal, interchangeable modules.
Established a design pattern for rapid prototyping of novel reinforcement learning architectures through imperative composition of the library’s base modules.
Built an end-to-end pipeline to train an autoencoder for real-time video compression, enabling robust communication for remotely operated vehicles in noisy environments.
Authored a custom C++ API using ExecuTorch to load and execute the highly-optimized inference model on resource-constrained microprocessors.
Modeling Portfolio Risk with Distributional Learning
Translated research for state-of-the-art reinforcement learning techniques into concrete implementations.
Developed a robust data pipeline to retrieve raw end-of-day trading data, process and structure raw data into useful representations, integrate new data into the existing database, and prepared the data for training.
Constructed a custom environment for trading simulation that realistically models market frictions.
Supervised and Unsupervised Learning for Word Relationships
Developed a proof-of-concept control system in C for a 3lb battlebot, integrating remote-controlled holonomic movement with an autonomous, lidar-based object-tracking subroutine.
Enabled autonomous orientation by interfacing Time-of-Flight sensors over I2C and writing logic to interpret ranging data for real-time rotational feedback.
Off-Grid Camper-Van Systems
Designed a stand-alone power system including multi-source charging (solar, shore, and alternator), automatic load balancing, routing, system controls and monitors, fuses, and safety shut-offs.
Designed water and propane systems, including storage, distribution, stove, water heater, sink, and shower.
Performed safety evaluations, sizing analyses, and cost estimation on system components.
Seamlessly integrated all systems into a fully functional and practical off-grid living solution.