Gautam Ramesh

· Explored multiple AI venture ideas, building MVPs and conducting user research to validate product-market fit. · Developed AI copilot for shipping brokers to match open vessels with cargo, answer pricing questions, and draft contracts; piloted with broker at a top maritime firm. · Built Tara AI, an AI tutoring app using spaced repetition that evaluates student responses and dynamically generates new practice questions; conducted user testing with∼10 students. React Native.

· Developed and launched several products in the AI entertainment space, including a 3D AI character with realistic lip-syncing and a copilot for writing and illustrating web comics. · Created a workflow that can take in up to four headshot images and generate pictures depicting all of them while adhering to the prompt - similar to what ChatGPT's image generator can do now, but released many months beforehand. · Reached a combined user base of ~10k across all products.

YouTube Ads Machine Learning team. · Developed deep neural network models using TensorFlow to predict skip and conversion rates for local business ads · Created a dashboard using SQL workflows to automate analysis of new model performance · Fine-tuned Google’s MUM large language model to detect deals in ad content, and built a pipeline to continuously generate a feature representing this for use in real-time ad selection models.

· Student-led organization sponsored by University of Pennsylvania to democratize access to Penn services by creating high-quality software. · Key contributor to Office Hours Queue (ohq.io), a remote office hours platform currently integrated into 100+ courses at Penn. Has facilitated 50k+ questions from thousands of students. · Technologies: JavaScript, TypeScript, React, Redux, Git

· Developed algorithm to transform the entire Amazon catalog into a snapshot with ~300 features per item · Worked with distributed datasets in Apache Spark and utilized functional programming techniques in Scala to optimize the runtime of the snapshot generation process · Snapshot fed into machine learning models that improve customer experience by identifying and grouping related catalog items together into a single listing with variation (size, color, etc.) dropdowns

· Graduate-level machine learning course covering support vector machines, principal component analysis, neural networks, deep learning, and more. · Conducted office hours and wrote/graded problem sets, exams, and Jupyter Notebook coding exercises.

· Scraped social media to identify issues that constituents care about and tailor messaging · Created elegant data visualizations to convey data-driven recommendations to non-technical campaign staff · Technologies: Python, R (ggplot, tidycensus, leaflet)

• Worked with students to deepen their understanding of proof-based discrete mathematics (including set theory, combinatorics, and graph theory). • Wrote and graded problem sets, conducted weekly office hours, and taught a weekly 10+ student recitation.