Portfolio

Mahima Shah

always learning

Scroll

About me

Data Science student at the University of Michigan who gravitates toward the intersection of technology and design.

I'm drawn to building things that are clear, thoughtful, and genuinely useful. I enjoy the process of taking an idea, understanding the people it's for, and shaping it into something that feels intuitive to use.

Outside of data and design, I love to travel and document the moments that shape how I see the world. I'm always experimenting with new crafts, from tactile ones like crocheting to digital projects on Procreate. If I'm not making something, I'm probably just catching up on a favorite series.

Mahima Shah
UI / UX Figma Design Systems Prototyping User Research Visual Design

Selected work

Projects

I build

01  /  Systems Engineering

Search Engine

A scalable search engine built from scratch — MapReduce pipeline over 28K Wikipedia documents, tf-idf + PageRank scoring, three parallel index servers, and a Flask search UI returning results in under 200ms.

Python MapReduce Flask REST API
Read about it
program 0.2818 00111573 3 325.75
michigan 0.4163 47914064 515 437.9
wolverine 1.7958 47914064 5 437.9
search 0.3241 00266386 12 182.4
index 0.5512 00350748 8 291.6
flask 0.8834 00121099 3 79.53
485
Search Engine

02  /  Full Stack Engineering

Instagram Clone

A full-stack social media app with a Flask REST API, React front end, infinite scroll, optimistic likes and comments, and double-click to like. Deployed to AWS EC2 and scaled to 3,000+ requests per hour.

React Flask REST API AWS
Read about it
Insta485

03  /  Machine Learning

ICU Mortality Prediction

An end-to-end classical ML pipeline that turns 12,000 messy, irregularly sampled ICU records into calibrated mortality predictions, using feature engineering, regularized logistic regression, kernel methods, and a bootstrapped evaluation with a held-out challenge submission.

Python scikit-learn Feature Engineering Kernel Methods
Read about it
0.81 AUROC

04  /  Machine Learning

Dog Breed Classification

Built and compared CNNs, transfer learning, and a Vision Transformer in PyTorch for dog breed classification, then designed a custom CNN with augmentation, label smoothing, and a cosine learning-rate schedule to close the validation-test generalization gap.

PyTorch CNNs Transfer Learning Vision Transformers
Read about it
10 Dog Breeds