I'm a computer science student at the University of Michigan, now pursuing a Master's in
Computer Science (expected May 2027) after finishing my B.S.E. this past summer. I've spent
time building things at the intersection of machine learning, cloud infrastructure, and
software engineering.
Most recently I interned at Google as a software engineering intern in Sunnyvale, working on
on-device machine learning. Before that, I did ML research in biomedical imaging — fine-tuning
vision-language models on CT scans to generate radiology reports.
I like problems that require you to think across levels of abstraction, from model
architecture down to cluster scheduling.
aside
I enjoy running, camping, reading, and watching the Knicks.
experience
work
Google · Sunnyvale, CA
- Built a low-latency, offline Android app in Kotlin and C++ via JNI running multi-model on-device inference pipelines — real-time audio streaming through a MediaPipe graph for voice cloning, dialogue orchestrated with Gemma in LiteRT-LM format, and speech synthesized by a neural TTS graph conditioned on the cloned voice.
- Accelerated on-device speech translation evaluations with a distributed pipeline using Flume to parallelize inference across multi-threaded workers, saving 200+ hours and catching quality regressions during LiteRT-LM runtime conversion.
- Extended the parallel evaluation framework to automatic speech recognition, demonstrating model performance on prospective client datasets in a 2-day turnaround for deals covering 15M devices and 800k vehicles.
U-M Biomedical & Clinical Informatics Lab · Ann Arbor, MI
- Fine-tuned the Qwen2.5 vision-language model on 50k abdominal CT scans to generate written radiology reports and classify abnormalities; built a pre-processing pipeline converting volumetric data to video.
- Cut training time 50% via LoRA, multi-GPU data parallelization with sharding, gradient accumulation and checkpointing, and quantization.
- Scheduled deep-learning jobs on an HPC cluster with Slurm; instrumented end-to-end monitoring with Databricks and Neptune.
Statistics Online Computational Resource · Ann Arbor, MI
- Worked in a team of 5 building a Virtual Hospital web app — designed cloud infrastructure using AWS S3 and AWS Fargate/Lambda for serverless compute.
- Built a modular integration pipeline for standalone R Shiny apps via Docker containerization, deploying 3+ analytics modules (e.g., data obfuscator, risk estimator) and cutting integration time by 70%.
- Reduced redundant user S3 buckets by 30% through a reworked cloud storage structure.
Aikito · New York, NY
- Designed BDD acceptance tests using Behave and Python Playwright as a gate before code commit.
- Built front-end features in React and TypeScript to streamline deal workflows between business owners and construction vendors.
projects
- Full-stack web app on AWS integrating the Charles Schwab API with OAuth 2.0 PKCE for real-time portfolio analysis — option position totals, visualizations, session handling.
- Built RESTful API endpoints and a Bootstrap interface for authentication and account analysis.
- React + Flask full-stack clone with REST APIs for auth, image uploads, likes, and comments backed by SQLite.
- SHA-512 password encryption; deployed as an AWS EC2 instance.
languages
C/C++, Python, JavaScript, TypeScript, Kotlin, Swift, R, MATLAB, Verilog, SQL, HTML/CSS
frameworks
PyTorch, Sklearn, React, Flask, Node.js, PostgreSQL, SQLite
tools
AWS, Docker, Slurm, Flume, Git, Jujutsu, Databricks, Neptune
bookshelf
currently reading
East of Eden
John Steinbeck
recents
A Passage to India
EM Forster
Ulysses
James Joyce
To the Lighthouse
Virginia Woolf
Burmese Days
George Orwell
Mrs Dalloway
Virginia Woolf
highlights
Something Happened
Joseph Heller
A Portrait of the Artist as a Young Man
James Joyce
Slaughterhouse-Five
Kurt Vonnegut
Still Life with Woodpecker
Tom Robbins
Moby-Dick
Herman Melville