Data Science Case Studies
Forecasting, evaluation, applied ML, and decision-support tooling built for messy real-world constraints.
This section covers the work where modeling and product meet: forecast quality systems, localization logic, tokenomics simulations, experimentation analysis, and interactive tools that help people think more clearly about probability and operational risk.
World of Warships Drop Simulator Case Study
An unofficial decision-support tool that turns drop-rate assumptions and pity rules into clear probability curves, expected values, and collection odds.
Read the ship-drop simulator case study
Football Localisation with Context Cartridges
A localization pattern that keeps generated sports content culturally native without letting the model invent the wrong teams, references, or tone.
Read the football localisation case study
Forecast Quality Beyond WAPE
A pragmatic evaluation system for deciding whether a model is safe to ship, not just numerically impressive.
Explore the forecast-quality framework
Post-Launch Smart-Contract Data Lessons
What the first launch cycle exposed about reward bugs, reporting gaps, and the data roadmap needed to keep the system legible.
Read the post-launch smart-contract lessons
Oracle-Node Penalty Design
A penalty-design exercise balancing operator incentives, fault tolerance, and the need to keep bad behavior economically unattractive.
Read the oracle-node penalty design case study
Investor-Unlock Scenario Simulation
A scenario simulator built to show how unlock design changes sell pressure, allocation dynamics, and governance conversations.
Explore the investor-unlock scenarios
How App-Store Rank Affects Sales
A mixed econometrics and feature-importance study on how app rankings, impressions, and sales actually interact.
Read the app-store ranking analysis
Advanced Forecasting with Prophet
A practical forecasting workflow for turning messy game-sales data into explainable predictions without building a bespoke model stack from scratch.
Read the Prophet forecasting workflow