# University Projects 🎓

These projects were developed during my time in the University of Michigan Masters of Applied Data Science (MADS) program. From demographic analyses tackling Detroit's digital divide to cryptocurrency clustering experiments and weather-dependent mobility studies, each assignment pushed me to apply cutting-edge data science techniques to real-world problems.

Whether working with geographic information systems or building interactive visualizations, these projects represent the foundation of skills I use today - from mastering statistical inference to communicating insights through compelling dashboards. Dive into the Streamlit apps below and explore how academic coursework transformed into practical data science practice!

# Detroit Digital Inclusion Project

This MIDAS initiative tackled a critical urban challenge: mapping digital access gaps in Detroit. Using demographic and infrastructure data, we identified neighborhoods most vulnerable to the "digital divide." This is essential for effective policy decisions around broadband expansion.

# Capstone: Finding the Next Big Thing

Our final project was a pipeline with the goal of finding the Next Big Thing. Freeform text like "Squid Game" was the input, resulting in finding the next new thing like "Squid Game" that was being talked about on Reddit. Wikipedia and Reddit APIs were used, along with the construction of a Conditional Random Field model to find unseen things, then a Top 10 list was returned.

Full interactive app and code: Streamlit demo | GitHub repo.

# Milestone II: Cryptocurrency Analysis

Dendrogram of coin clusters based on price movement

Course 695 challenged us to combine supervised and unsupervised learning approaches. My project explored cryptocurrency markets from two angles:

Supervised: Next-day price prediction using historical patterns (a practical, if ambitious, forecasting exercise).

Unsupervised: Discovering "crypto families" - groups of coins whose prices move together. Using dynamic time warping to measure similarity in z-scored price trajectories despite different absolute values, then clustering coins into communities with shared behavior.

The resulting dendrogram reveals interesting market microstructure: smaller altcoins often cluster by ecosystem or use case, while large caps like BTC/ETH form their own independent clusters apart from other large coins. Full interactive app and code: Streamlit demo | GitHub repo.

# Milestone I: Rain & Taxi Mobility

Course 592 introduced us to difference-in-differences (DiD) - a causal inference workhorse in economics and social science. My first DiD application measured how weather affects NYC urban mobility.

By comparing pre/post rain periods while controlling for time trends, the analysis isolates the causal impact of precipitation on taxi demand. The results show that New Yorkers adapt their travel behavior to weather conditions by taking more taxi rides when there is precipitation compared to a hypothetical dry-day baseline.

Explore the visualization to see how different weather intensities shift mobility patterns. Complete notebook and Streamlit app: GitHub repo | Interactive demo.

# Test Deepnote to Streamlit

A lightweight demonstration of modern data science workflows: cloud-based coding in Deepnote notebooks, version control via GitHub, and instant deployment as a Streamlit web app.

This setup shows how you can collaborate on data projects online while maintaining reproducibility. The included templates demonstrate convenient Deepnote features that streamline the development process.

Want to try it yourself? View the live demo | GitHub repository.