The Coca-Cola Company, summer 2026
~50people used the tool I built
A four-person, three-day manual pull became an internal tool. Azure OpenAI plans the query, plain Python computes every figure, and 50+ checks stop the run if two systems disagree.
Case study
Quadcore Innovations, 2025 to 2026
100 to 40support tickets after the fixes I proposed
Traced about 70% of tickets on a payments platform to one crash path. Resolution time went from 2 to 4 days to 24 hours, and weekly reporting moved to automated SQL feeds.
Case study
UW Bothell Club Council, 2026 to 2027
Reviewing a $200K+ student budget
nearly 20%of funds allocated so far needed a correction I made before approval
As Treasurer I check each club’s purchase request against its quote, invoice or cart, line by line, before it goes to a vote. The allocation covers 95+ organizations.
Team Cartier, team project
Forecast vs actual P&L
25+expense categories broken out
A FY2024 wholesale-distribution P&L: forecast vs actual by channel, budget variance and working capital. I built the P&L waterfall and broke out the expense categories. Figures are the case dataset.
Dashboard
Independent research, 2026
Who really gets aerospace contract dollars
$499.4Bof DoD aerospace spend, FY2021 to 2025
Every NASA contract transaction plus DoD aerospace manufacturing (NAICS 3364) from USAspending, rolled up to parent companies. Lockheed Martin’s share goes from 21% as listed to 43% rolled up, and 87% of DoD aerospace dollars were not competed.
Live findingsCode
Independent research, 2026
Three-way match audit of a real ERP log
3,693PO items invoiced before the PO existed
Fourteen purchase-to-pay controls run on a public log of 251,734 PO items. 1.47% of items had an invoice dated before the PO, and 2,526 had one person both record and clear the same invoice.
Run the auditCode
Independent research, 2026
Testing the bullwhip effect
44industries tested on Census data through 2025
A replication of a well-known bullwhip study. With seasonality kept, no retail industry amplifies demand. Remove seasonality and 82% of retail cells appear to, so the method can manufacture the effect.
Memo and chartsCode
Independent build, 2026
Seeing Boeing’s production rate from the sky
624 of 774737 deliveries seen as first flights
A 3D Puget Sound where every new Boeing jet’s first flight replays from public flight-tracking history, with a ledger that compares the sky count with Boeing’s reported deliveries over 647 days.
Explore itCode
Independent build, 2026
Satellites as a supply chain
16,896active satellites in the feed at last check
Every active satellite over you, with Amazon Leo and Starlink shown as a production line: factory, launch, orbit raising, on station, retired. Both factories are a short drive from campus.
Look upCode
Independent research, 2026
4,433settled markets scored
A temperature model scored against Kalshi prices. The market was better calibrated (Brier 0.2114 vs my 0.3047), so I did not trade real money on it.
Case study
SQL study, 2026
54.4%of revenue from a third of buyers
Cohort and window-function queries on a six-table retail schema with 620 orders. Customers with four or more orders are 33.3% of buyers.
Case study
Independent build
Economic Pulse dashboard
7live indicators from FRED
Seven live indicators from the Federal Reserve’s FRED API in Python and Streamlit, with recessions shaded.
DashboardCode
Independent study
Washington housing and homelessness
9years of rent, income and vacancy data
Nine years of rent, income, vacancy and homelessness data, modeled in R and published in Tableau.
ReportCode
Open source
ragproof
BEIRSciFact benchmark harness
An open-source harness that scores how well a search system finds the right documents.
Code
Product I built
AutoAppli
~400weekly users
A job-search tracker that shows where an application pipeline is leaking.
Site
Product I ran, Mar to Aug 2026
SyllabusAI
1,055automated tests
Turned a syllabus into dated calendar deadlines. I built and ran it from March to August 2026.
Site
Kaggle simulation competition, 2026
A card-game playing agent
245th of 6,807final rank, top 3.6%
An agent that plays the Pokémon Trading Card Game, ranked on a ladder against thousands of other entrants’ agents.