• Implemented a web-scraper and search API for Facebook, LinkedIn, and 3rd party websites to accumulate sentiment data on the almond industry

• Developed and pushed a persistent full-stack Flask Plotly Dash dashboard to production citing thorough documentation of model result comparisons including ranked features, chosen model frequency and yield prediction variations for over 300 production runs within current harvest

• Improved data modeling robustness by minimizing impact of overfitting and noise through separating bimodal distributions and Hamiltonian Markov Chain Monte Carlo trace plots with 6000 iterations and 500 burn-in iterations

• Added a minimally correlated feature set based on VIF threshold that improved mean percent error by 70% for poorly performing models

• Working on Latent Repository State NLP modeling with unsupervised learning on multimodal datasets

• Implementing β-VAEs and Wasserstein GAN-BERTs for malware detection on open-source data such as Twitter, GitHub, Stack Overflow, and YouTube

• Allow for actionable representation of the state to fill in missing information while detecting deviations in user and moderator activity

• Worked with Security, Detection, and Automation Dpt. and TSA in creating and analyzing models determining if a certain article contains explosives

• Trained XGBoost binary classification models with 23000 training and 30000 test images, scored based on Detection and False Alarm Probability

• Improved hit rate from 35% to 60% and decreased false alarm rate from 3% to 1% a dataset with 93% negatives and 7% positives.

• Transferred existing codebase into Dataiku DSS, allowing for a 200-personteam to have a more efficient pipeline and promote readability

• Implemented and trained object detection CNNs based on 10000 X-ray images of cargo truck side profiles containing various manifests

• Revamped legacy models to work with updated packages and to fit the company’s code standard

• Used linear regression, random forest, hierarchical Bayesian models, and other statistical modeling on data from GEE and leverage customer data to predict the acreage of almonds for over 5000 agriculturalists across 3 different states

• Incorporated 80 features including the quality of soil, weather patterns, and irrigation methods in the predictive models from data warehouse (AWS)

• Implemented feature clustering and sub clustering to investigate 30 highly correlated features using statistics metrics like Variance Inflation Factor