varchi [at] seas [dot] upenn [dot] edu
I'm currently a student at the University of
Pennsylvania pursuing a BSE and MSE in Computer Science with a concentration in
AI. My interests
span
distributed systems,
machine learning, optimization, web/mobile development, and more.
I am also a D1 athlete on the varsity
squash
team.
In my free time, I like to do photography.
Relevant Coursework
- CIS 1200: Programming Languages and Techniques
- CIS 1600: Discrete Mathematics
- CIS 1210: Data Structures and Algorithms
- CIS 2400: Computer Systems
- CIS 2620: Automata Computability, and Complexity
- CIS 5200: Machine Learning
- CIS 5300: Natural Language Processing
- ESE 2000: AI Lab
- ESE 4020: Statistics for Data Science
- ESE 5140: Graph Neural Networks
- MATH 1400: Single-Variable Calculus
- MATH 1410: Multivariable and Vector Calculus
- MATH 2400: Linear Algebra and Differential Equations
Work Experience
May 2024-Present
Philadelphia, PA
- Developed generative image watermarking architectures based off CNNs and diffusion models to synthesize images with hidden and robustly embedded watermarks.
- Implemented generative watermarking models such as Stable
Signature and Recipe for Watermarking
Diffusion Models
using a
new constrained objective with dual learning as described in PACC
Learning.
- Conducted in-depth analysis between watermarking models with different architectures varying loss constraints, dual learning rates, and other hyper parameters. Model performance was tracked during training using WandB.ai.
May 2024-Present
Philadelphia, PA
- Independently devised comprehensive solutions for machine learning labs,
including Variational Autoencoders, Variational Diffusion Models,
Actor-Critic Reinforcement Learning, and Large Language Models.
- Authored in-depth explanations accompanying solutions, ensuring clarity
and rigor by integrating detailed mathematical proofs.
Some explanations were published in notes.
July 2022-May 2023
Madison, WI
- Managed cloud infrastructure, CI/CD pipelines, and DevOps services
for datachat.ai,
optimizing
deployment processes
and
enhancing
system reliability.
- Designed a novel microservice architecture to handle machine learning tasks concurrently, improving system scalability and reducing request dead locks.
- Transferred our entire AWS Kubernetes deployment from UI-based to IAC (Infrastructure as Code) using Terraform, while automating infrastructure updates through CircleCI, Docker Compose, and AWS.
Projects
Generative Diffusion Model for Video
- In collaboration with researchers from Google Deepmind and ASU,
helped develop a generative diffusion model for video using the DDPM
algorithm, implementing it from scratch using Hugging Face UNets.
- Automated data collection in Unreal Engine using the Unreal Python API, generating 2D projections, depth maps, and normal maps to create a comprehensive dataset for model training.
- Created Rift, an iOS application built with SwiftUI, offering students
an enhanced in-app experience for managing their academic journey.
- By reverse-engineering the Infinite Campus API, Rift provides students with seamless access to vital information such as grades, homework assignments, and messages, all within a user-friendly interface.
- Currently being maintained for over 15,000 middle and high school
students.
- Designed, developed, and launched easeattendance.com, a web
platform
designed to seamlessly integrate with Zoom meetings, streamlining the way educators manage attendance.
- The platform offers real-time attendance insights supported through
websockets and can store detailed statistics for students via
distributed NoSQL databases.
- AES-256 encryption to encrypt all student records in our
databases in order to comply with FERPA data privacy requirements and make our product available to all US schools.
- Developed a headless server for automated options day trading,
integrating a TradingView-hosted algorithm with the Ameritrade API. Utilized RSI (Relative Strength Index) to optimize trading positions.
- Performed backtesting to check algorithm performance. Realized a 30% gain in 3 months of using the algorithm.
Research
- Co-authored a research paper focused on leveraging deep learning
computer vision techniques for predicting poverty levels in various global regions using overhead satellite imagery.
- Addressed the challenge of limited reliable economic data in developing areas, offering a more cost-effective alternative to traditional surveys.
- Explored the impact of data quantity and augmentation on network performance in order to improve the model.
Awards
- Competed in USA Computing Olympiad coding contests and reached the
gold division.