• AI Core Modeling Team

• Software Engineering Intern on Two Sigma's AI Core Team

• Coordinated a team of 11 course staff to run CS 189: Intro to Machine Learning (150+ students)• Taught weekly sections on machine learning concepts such as likelihood estimation, SVMs, and attention mechanisms• Developed course content such as biweekly homework problem sets, skeleton code for programming assignments, and exam questions• Managed course website (eecs189.org)

• Researched efficient NLP for edge computing and real-time text generation, advised by Amir Gholami, Sehoon Kim, and Professor Kurt Keutzer• Worked to improve the latency and parameterization of Gmail’s smart compose model via pruning, quantization, and knowledge distillation

• Software Engineering Intern on Facebook's Ads ML Modeling Team• Proposed a novel optimizer setting for Facebook’s Ads Ranking ML Algorithm, which is now implemented in the latest productionized model because it significantly improved model performance• Investigate the effectiveness of different optimizers on Facebook's Ads Ranking machine learning algorithms via experimentation on various Caffe2 models• Write python scripts to analyze model performance, learning curves, and tune hyperparameters

• Developed a CycleGAN Neural Network for image-to-image translation• Performed data preprocessing and neural network training for Percipient.ai and SiMa.ai• Completed training in the usage of frameworks such as Scikit-learn, TensorFlow, Keras, Pandas, Numpy to implement convolutional neural networks, support vector machines, etc.

• Organized a semester-long crash course on machine learning concepts for new members of the organization• Taught new members about machine learning topics including SVMs, CNNs, data preprocessing, etc.

• Benchmarked 12 ML models on ImageNet and Coco datasets to analyze power statistics on 3 different GPUs (Models include ResNet-101, Inception V2, VGG 16, SSD, etc.)• Wrote python and bash scripts to leverage NVidia TensorRT Framework on AWS instances to optimize machine learning architecture via Int8 quantization

• Organized guest speakers events to provide M.E.T. students with career advice and information on various career paths, such as Quantitative Trading and Private Equity• Worked with fellow M.E.T. Student Board members to organize Cal Day Events, introductions to current students, and informative posts regarding life as an M.E.T. student in order to encourage high school admits to join the Berkeley M.E.T. program• Planned and hosted social events to foster M.E.T.'s tight-knit student community, such as student brunches and retreat• Actively monitored and sought to improve the M.E.T. internship recruiting pipeline

• Collaborated with Percipient.ai to use neural architecture search (NAS) to optimize computer vision tasks• Preprocessed and cleaned satellite data on vehicle detection for our neural architecture search task• Worked with team to fully implement recently published RL and gradient-based NAS research papers (SNAS & DARTS+)• DARTS achieved a 13% increase in validation accuracy over our baseline ResNet-50 model on the given task

• Oracle NetSuite is the #1 rated Cloud ERP System in the United States• Worked on implementing a random forest machine learning algorithm for business predictions in NetSuite software• Developed new features in NetSuite enterprise software for business finance and accounting using Java and SQL, such as a tool to update multiple items' inventories in parallel, a generic data accessor architecture to integrate machine learning into NetSuite's software, and more• Fixed numerous issues in NetSuite's latest release version in order to satisfy customer needs

• AI firm that predicted the 2018 Oscars awards with 94% accuracy and the Kentucky Derby superfecta (541-to-1 odds)• Analyzed Swarm AI input data to develop metrics to quantify anomalous behavior in swarms• Developed an unsupervised learning model for anomaly detection in swarm datasets, which identifies anomalous behavior with 92% accuracy

• Researched in the Bioengineering Department under Dr. Adele Doyle• Investigated the influence of physical cues on stem cell lineage, created computational model of protein pathways in a stem cell to explore novel effects of physical stimuli on stem cell lineage• Discovered a statistically significant correlation between the protein concentrations of the YAP/TAZ protein and stem cell lineage affected by physical stimuli, indicating that YAP/TAZ is a significant protein for regulating the gene expression that controls stem cell differentiation."YAP/TAZ and ß-Catenin Interactions Directing Stem Cell Lineage: A Computational Modeling Approach"• Presented at UCSB RMP Symposium• 2017 Siemens Competition Semifinalist