● Buiding data-driven solutions to clients worldwide● Worked with companies of all sizes, from pre-seed startups, to VC-funded startups and fortune-100 enterprises

● Developed a model to optimize bids on auction-based online marketplaces for a top consumer food company. The model combined a CatBoost tree ensemble and a Bayesian model to predict sales from marketing spending. Improved net revenue on Kroger by 20% on a 6-week-long A/B test.● Designed a Bayesian diff-in-diff test for A/B testing based on an in-house python package for Bayesian tests. Conducted A/B tests, from finding testing pairs that matched criteria stated by the business to monitoring status and summarizing results.● Collaborated with ML team to productionize bid suggestion model, writing a clean software package with > 85% test coverage, concize configuration files, and containerization for CI/CD. The production-ready version is now running as Kubeflow dags.

• Designed neural network architectures for object detection and semantic segmentation in the context of seal detection in high-resolution satellite imagery.• Created an ensemble approach for seal detection using CatBoost tree-based model to combine outputs from multiple CNNs into consensus predictions, outperforming human observers at seal detection.• Applied similar techniques to several use cases in computer vision including penguin colony size estimation and sea ice segmentation in satellite imagery and whale detection in aerial imagery.• Awarded twice through the Stony Brook Institute of Advanced Computational Science Junior Researcher Fellowship.• Employed an array of custom-designed object detection convolutional neural networks empowered by NSF HPC machines to process a 500TB archive of high-resolution satellite imagery detecting seals.• Published results as several publications in high-impact journals and conferences, including Remote Sensing of Environment, CVPR, and Remote Sensing.

• Saved 100s of UGTA hours by writing python scripts to organize grading sheets for large intro BIO courses• Conducted several lab classes and lectured groups of over 200 students• Mentored advanced bio students on landscape ecology and applied ecology labs including GIS lessons

• Designed and developed an anomaly detection pipeline using a combination of Isolation Forests and population statistics from historical averages, comparing and contrasting the most unusual data points with the most typical data points.• Implemented a feature drift validation pipeline that flags anomalous features using the Kullback-Leibler divergence from a past baseline as a criterion within Great Expectations.• Calculated probabilities for Reinforcement Learning model recommendations for different RL agent types and exploration strategies to test agents/exploration strategies on past data using importance re-sampling.

• Developed computer vision pipelines for whale detection in satellite imagery for clients in the environmental and oil sector.• Conducted experiments with customized loss functions, synthetic imagery, and data augmentation to improve the potential of training datasets.• Developed code functionality to train, validate and test semantic segmentation deep learning models for whale detection.

● Developed an ML pipeline with LightGBM to detect pregnant users based on their interaction with the healthcare plan.● Working on feature engineering to detect patients with chronic diseases and audit fraudulent medical bills from a diverse portfolio of > 600,000 lives and a decade-long usage dataset. ● Assisting with the transition to cloud computing with Oracle Cloud Infrastructure.

● Built 'Birds of a Feather', a birding partner recommender system● Compiled data from eBird observation records (> 300gb of records, 15 years of spatialized observations) for active users (> 100k users) using 25 hand-engineered features.● Built a recommender system by training a Siamese neural network to encode 'birding style' and match similar users using cosine distance between encoded features.● Deployed Interactive Web App on AWS ec2.● Distilled the project into a concise and captivating 5-minute pitch and presented at Insight partner companies including Bloomberg, Via, and ChartBeat.