https://storycrafter.ai

https://mlestimation.com

Discovered issues of algorithmic bias by showing that a small change in the framing of labeling question prompts can result in a significant distribution shift in labels, which propagates to affect downstream model predictions.• Designed web-based surveys using A/B testing to assess the contentiousness of different clothing attributes when included in a dress code.• Collected labelling data for clothing attributes of 4,000 images using Amazon Mechanical Turk. • Created dress code violation classification models achieving over 90% accuracy and extended this framework to pet codes for apartment building and meal policies for schools.

• Developed a novel meta-learning approach for cancer therapy benchmarking to distinguish between good and bad treatment plans.• Conducted experiments on clinical data and demonstrated that meta-learning models trained using MAML or Reptile algorithms can outperform the previous isolation forest model by up to 15% for certain cancer types. • Investigated methods for data representation of cancer therapy treatment plan data by using feature learning approaches involving neural networks, clustering, and principal component analysis.• Presented research methodology and results at the Google Accelerate program for startups.

• Facilitated continuous integration and development of software by creating configurations in QuickBuild holding the information needed to generate builds.• Enabled flexibility in updating and adding new configurations by implementing configurations in a hierarchical structure, allowing for more efficient development, testing, and production of software.

• Automated a data pipeline that pulls semiconductor wafer test results from the database to generate weekly yield reports.• Eliminated the need to manually select wafers for profiling by creating a python program to cluster and select the required wafers based on parameters extracted from wafer acceptance tests.• Expedited the adaptive voltage frequency scaling process by prototyping an algorithm in python to predict critical path frequencies and generate hex encodings for path masking to be used in the firmware.• Led training sessions to promote the use of python scripting to product engineers in multiple offices (Austin - Texas, Singapore, Shanghai - China).

Designed and implemented a user-friendly GUI for TCT application.Created a personal website for my Professor.

Instructed middle school students in essay writing, and helped develop their oral communication skills.