a dask cluster on ec2 crunching amazon product and review data that doesn't fit on one machine: distributed dataframes, partition-aware joins, lazy evaluation, and a lot of debugging memory pressure.

combined u.s. outage, economic and climate data to predict which outages turn severe, then checked whether the model treats low- and high-income regions fairly.

paste a github link and get a plain-english tour of the codebase. a full-stack rag app that embeds the repo, searches it, and has an llm answer with real context.

a hybrid fraud score that blends gradient boosting with an autoencoder's anomaly signal, evaluated the honest way for a heavily imbalanced problem.

a scroll-driven d3 story about california wildfires, built on cal fire data from the last decade.

a working router with arp handling, ttl decrementing, longest-prefix matching, and custom firewall rules controlling traffic between hosts.

a crawler built from scratch: a crawl frontier, url deduplication with hash tables, and page storage, all over libcurl.

models that predict which shelter animals get adopted quickly, plus a fairness check across animal groups and a full statistical testing pipeline.

nlp on 124k+ linkedin job postings to predict a role's category and seniority from its description, with a streamlit demo on top.