- Researched and prototyped various pre-trained models (GPT, BERT variants) to determine the most suitable for AI-assisted email generation- Developed and refined prompt engineering strategies, including designing prompts that provide sufficient context to generate relevant and accurate responses, as well as managed prompt chaining for handling more complex queries- Managed the entire data and model pipeline from development to deployment- Implemented basic personalization features, including feedback loops which enabled the platform to learn from user preferences and interaction

- Designed and implemented a robust set of interview questions and case studies to effectively assess candidates' proficiency in key data science concepts, programming languages (Python), and industry-standard tools (SQL, TensorFlow, scikit-learn)- Partnered with hiring managers to optimize the technical interview process while ensuring a thorough and fair evaluation of candidates’ problem-solving abilities and technical understanding- Delivered comprehensive feedback on candidate performance, emphasizing strengths and areas for growth, which directly influenced hiring decisions and enhanced the efficiency of the recruitment process

- Conducted in-depth research and exploration of AI models to address intricate challenges within the voluntary carbon market- Spearheaded the implementation of AI-powered pricing solutions for carbon credits, leveraging TensorFlow and Keras to achieve a 27% increase in accuracy compared to previous methods- Devised unique similarity measurement strategies to address complex data patterns- Engineered recommender solutions that improved system performance and reliability in data-scarce environments- Successfully enabled consumer interaction with models by deploying production-ready modules accessible through our API and web UI, increasing user engagement- Created and managed daily ETL workflows, ensuring timely and accurate data delivery for the market team's analytical needs- Leveraged Apache Airflow to automate the entire model lifecycle, from data extraction to model evaluation and retraining

- Designed and implemented a robust rules engine to transform data received in over ~100 disparate formats into a standardized and actionable format- Designed and managed a robust relational database, incorporating a user-friendly interface, to facilitate seamless data management for historical, current, and upcoming datasets- Engineered and automated end-to-end ETL processes to cleanse raw data and efficiently load structured data into a SQL database, resulting in a substantial reduction of receipt-to-load time from days to minutes- Designed and implemented a robust relational database and intuitive user front-end, facilitating secure remote access, utilization, and management of electronic documents for payroll processing and customer service teams during the Covid-19 pandemic, effectively mitigating the risk of exposure and illness

Phenomenology of high energy collisions of protons relevant to the Large Hadron Collider- Developed a mathematical formalism to satisfy and preserve constraints on parton momentum distributions, allowing for numerical computation of the cross section for the Drell-Yan process while accounting for double-parton interactions.- Used MATLAB, Mathematica, and C++ to analyze the multi-parton interactions from a theoretical and phenomenological standpoint using existing data.- Coauthored "Constraining the double gluon distribution by the single gluon distribution", published in Physics Letters B.

I worked to analyze and distinguish atmospheric neutrino signals in hopes of being able to differentiate them from proton decay signals.- Created a 3D particle interaction simulation in C++.- Included a complete range of functionality including tracking, geometry, physics models, and hits.- Ran spatial statistics to analyze distances between particle interactions in ice, chlorine, and gadolinium.- Ran signal statistics to differentiate between event signal and background events.- Performed cost analysis on a real world implementation of this detector and concluded real world implementation would be too costly.