A backend for flash sales, where thousands of people try to buy limited stock at the same moment. Orders go through a Go/gRPC gateway into a Redpanda queue, so the database never takes the spike directly, and stock is claimed with atomic Redis scripts so nothing gets sold twice. Under a k6 load test it handled 100,000+ requests without dropping any, even with workers crashing and the database down for 20 seconds.
Hi, I'm Shreepaad. I'm a math and computer science student at UIUC, graduating in 2028.
Most of what I build is backend and systems work. This past summer I was at Northrop Grumman working on a mission computer simulator, and before that I did algorithms research at Northwestern. On the side I build things like a queue that holds up under flash-sale traffic and a job application autofiller that runs entirely on your own laptop.
Experience
Northrop GrummanSoftware Engineering Intern · Rolling Meadows, IL
Worked on a C++/CLI simulator for an aircraft mission computer. I rewrote how it sends and buffers messages so it could talk to newer flight software, and wrote multi-threaded C++ and Python test clients that cut manual testing time by about 40%.
HackIllinoisSoftware Engineer · Champaign, IL
Built parts of the site and tooling for the Midwest's largest hackathon in React and TypeScript, on MongoDB and S3, used by 2,000+ people. I also fixed a state management problem that made the UI thrash, which made rendering about 65% faster.
Northwestern UniversityAlgorithm Research Intern · Evanston, IL
Researched task scheduling algorithms with Prof. Khuller. I built a scheduler around Maximum Weighted Independent Set (pick the most valuable set of tasks that don't conflict), which cut runtime by up to 90% across the scenarios we tested.
Illinois Institute of TechnologyMachine Learning Intern · Chicago, IL
Worked with Prof. Cheng on adding wavelet transforms to graph convolutional networks. The model versions I helped test improved prediction accuracy by over 16%.
Projects
A Chrome extension that fills in job applications using a language model running locally through Ollama, so your personal details never leave your computer. It matches form questions to your answers with embeddings and keyword scoring, and leaves a field blank rather than guess when it isn't sure. It recognizes 117 kinds of questions.
Paste in a meeting transcript and get back a list of tasks, each with an owner and a due date. A classifier finds the sentences that are actually tasks (98% accurate), spaCy and dateparser pull out who and when, and a small model (phi-4-mini) writes a short title for each.
Skills
Languages
- C/C++
- Python
- TypeScript
- JavaScript
- Go
- Java
- R
- HTML/CSS
Systems & web
- Linux
- Docker
- Git
- CI/CD
- gRPC
- Redis
- PostgreSQL
- MongoDB
- AWS S3
- Redpanda
- k6
- React
- Next.js
- Node.js
- Flask
- FastAPI
- Firebase
- Chrome
Machine learning
- PyTorch
- TensorFlow
- Hugging Face
- scikit-learn
- pandas
- NumPy
- Ollama
- spaCy
Education
University of Illinois Urbana-Champaign
B.S. Mathematics & Computer Science · GPA 3.88