Working with data and dealing with problems I actually have.
New York City, NY
Built a lead-scoring pipeline that mashes together Google Places, PPP loan, and FEMA disaster data with Python and DuckDB. The result: 32,000+ ranked business leads, with the scoring rules sanity-checked against real customers.
Concentration in data science. Graduating December 2027.
A digital nature magazine on Instagram that grew to 130K+ followers and 150M+ impressions in five months. Engagement data (reach, saves, watch time) picked the subjects and posting times, and growth hit 200–400% week over week at its peak. Sold to a crypto company that ran it into the ground lol.
Collected COVID-19 data across NYC and supervised the collection, cleaning, and organization of 200,000+ data points from 25+ communities. Consistent logging steps kept entry errors down, and plenty of the job was talking with residents about sensitive health info.
A resale business that peaked at $80,000+ in annual revenue. Sales and pricing data decided what to buy, sell-through tracking showed what worked, and margins had to stay ahead of $900+ in monthly overhead.
Zicklin School of Business. Coursework in big data, Python and R for analytics, data warehousing, and business statistics.
Organized and categorized 7,000+ product samples for the annual sample sale, plus cleaned product data for marketing reports.
Helped run monthly community workshops on policy, campaigns, and social issues, teaming up with local activist groups and fundraising with communities in Queens and Brooklyn.
Hello all! I'm a Tibetan data science grad student in New York City, earning my M.S. in Analytics at Georgia Tech (expected Dec 2027). This past summer I was an AI intern at KarmaStaff, where I built a lead-scoring pipeline that pulled public data (PPP loans, FEMA disasters, Google Places) into a single tool that ranks prospects by how likely they are to need the service. Before that, I ran my own eBay reselling business, collected data in the field, and built web and data projects. Lately I've been working on causal inference and evaluating how well retrieval (RAG) systems find the right answer. Looking for data science roles in NYC or remote!
A timed game for practicing SQL. The clock starts at 60 seconds, every correct answer adds 10 more, and the questions get harder as you go.
Answers are checked by comparing normalized query text to a saved solution, not by running it against a database, so a clever query with the right result can still get marked wrong.
JavaScript · React · Monaco Editor · SQL
Steam, Discord, and social media kept winning the fight for attention during work time. This Windows app is the one-button fix: it blocks chosen websites and programs for a free or timed focus session.
The retro Windows look felt right for the job. It’s a focus tool, not a fortress: admin access or a VPN gets around it.
Python · PySide6 · Windows Firewall
A Chrome extension for digging through a YouTube channel’s videos. A collapsible toolbar filters by title, duration, views, and date, with sorting and saved preferences.
It only sees the videos YouTube has already loaded, so it can’t search a whole channel at once, and dates are estimated from labels like “two months ago.”
JavaScript · CSS · Chrome Extensions
The groundwork for fraud detection: join transaction and identity records, check missing values, and drop sparse or near-constant columns. The cleaned training set has 590,540 rows, and only about 3.5% are fraud (hello, class imbalance).
A replay script streams transactions at 10 per second. For now it just logs them; the model isn’t wired into the stream yet, and there’s no validated detection score.
Python · pandas · scikit-learn · XGBoost
How do fares, trip lengths, and tips work in New York? PySpark reads trip files from Amazon S3, handles missing values and outliers, and fits fare and tipping models on AWS EMR.
The data had negative trip distances, so cleaning mattered a lot. The published script covers six months of yellow taxi records (Jan–Jun 2021); the full TLC archive goes back 13+ years.
Python · PySpark · Amazon S3 · AWS EMR
Where did the PPP loan money go, and which lenders handled it in New York and New Jersey? The SBA’s CSVs get modeled for a warehouse, loaded into Google BigQuery with Talend, then analyzed in Python.
A crash course in moving data from local files into a cloud warehouse. The cleaning and analysis are public, but running the full project needs BigQuery warehouse access.
Python · SQL · Google BigQuery · Talend
Talks, deep dives, and rabbit holes worth the watch.
Jane Street ·
Justin Sung ·
Visual Kernel ·
MIT Siegel Family Quest for Intelligence ·
PawelCodeStuff ·
stoney codes ·
Philipp Lackner ·
Dwarkesh Patel ·
Talks at Google ·
Kaggle ·
Dwarkesh Patel ·
Jane Street ·