Alpha School Content API• Created an API to deliver articles and reading comprehension questions for Alpha School’s reading app.• Developed an LLM-based quality control system to ensure high-quality article content and questions.• Implemented a patching system to re-write low-quality questions, answer choices, and explanations.• Sourced literature content for all grades and SAT content for grades 9-12.• Created a system for article content-based question generation targeted to specific Common Corestandards (grades 3-8) and SAT skills (grades 9-12).• Prototyped “BothSides” generation of novel articles representing the opposite view of a political article.• Developed AI image generation and quality control for articles lacking or with low-quality images.• Tracked metrics and continuously delivered improvements to the student experience on the reading app Athlete AI• Developed an audio-based algorithm for high-energy moment detection.• Implemented an applause/cheering detection model using OpenAI Whisper.• Fine-tuned a YOLO object detection model for made basket detection.• Connected the object detection model with a MMPose pose model for dunk detection.• Employed a YOLO segmentation model and object tracking in a player team classification algorithm.• Integrated the highlight detection techniques into a cohesive basketball highlight detection system.Software Engineering/Machine Learning Project, AE Studio Neurotech Development Kit• Implemented conversion of 3D results to 2D in AE's open-source Neurotech Development Kit library.• Trained machine learning surrogate models to speed up focused ultrasound steady-state simulations.Other Projects• AE Hackathon: AI-generated crossword API• AE Hackathon: Align My AI website• Alpha School Post-Test Coach Bot• Analyzing Neural Time Series Data Python translation• NextSense in-ear EEG earbuds feature analysis/prioritization• Text-to-SQL bot• AE Hackathon: basketball play generator website

Contributed to the development of a functional ultrasound-based decoder for a real-time, minimally invasive brain-computer interface (BCI). Determined the minimum sized ultrasound volume needed for Power Doppler beamforming without affecting decoder accuracy for the purpose of increasing decoding speed. Reduced necessary image size to 10% of original size for a 90% reduction in Power Doppler beamforming time, key for the development of a real-time 3D ultrasonic BCI.

Created a GPU-based beamformer for 3D ultrasound imaging. Developed code in CUDA to process raw ultrasound data acquired at ultrafast frame rate with a matrix array probe into 3D images for 3D functional ultrasound imaging. Validated beamformer with in vitro and in vivo (rat) functional imaging experiments. Contributed to a 3D functional imaging sequence to acquire data at a gold-standard imaging frame rate of 0.5 Hz for functional brain experiments.

Trained reinforcement learning (RL) and recurrent neural network (RNN) models to predict decision making behavior based upon a self-reported depression score. Extracted behavioral differences between clinical populations from the RNN models to create a behavior-based mental disorder classification system.

Replicated the underlying generalized linear mixed model of the ACS NSQIP Surgical Risk Calculator. Developed novel surgery predictive models based on Random Forest and Fuzzy Forest models. Showed tree-based models led to no significant increase in accuracy over linear models. Improved the predictions of risk in surgery over the NSQIP risk calculator by an average of 14% using feature selection techniques.

Investigated centromere and kinetochore and their function in helping correctly segregate chromosomes during cell division. Isolated the Mis18 and Mis18 proteins of the Mis18 complex to learn more about their function in CENP-A recruitment to the centromere region.

Studied the molecular biology of Histoplasma capsulatum and its two temperature-dependent forms. Investigated the potential correlation between the makeup of a fungus and its growth temperature. Cleaned and visualized a dataset of maximum growth temperatures. Analyzed the data to find a predictor of temperature based on genome makeup.

Rootbound Rooms is an escape room created by six students from The Nueva High School, in San Mateo, California. We designed and created an escape room for the Nueva High School's Quest Night.

Collaborated with a group of software engineering interns to create a program that would allow beginner programmers to focus on learning programing concepts rather than having to worry about the syntax of specific programming languages. Created a “natural” programming language so first time programmers could code programs with natural language and put their focus on the concepts used.