Product / Data Science / 2026
Business Empire
I am building Business Empire, a strategy board game with a growing card pool, seeded Monte Carlo simulations, an interpretable ML balance model, and a playable Expo prototype.
Strategy board-game product with a 93-card simulation and ML balance pipeline, Expo prototype, and physical manufacturing direction.
- Role
- Game design, data science pipeline, Expo prototype
- Status
- In progress
- Tools
- PythonMonte Carlo simulationMachine learningPandasNumPyNetworkXStreamlitExpoReact Native
Story
I am building Business Empire as a strategy board game. I wanted the balance work to be more than intuition, so I used my Data Science final project to build a reproducible way to parse cards, simulate matchups, train a small ML surrogate, evaluate board layouts, and test small changes before physical playtests.
Build
- Parsed the current 93-card data set into structured game features for cards, industries, costs, income, tiers, and board-related behavior.
- Built seeded Monte Carlo tournament scripts that compare strategy profiles across repeated matchups and report win rates, balance score, card usage, and board telemetry.
- Trained a ridge-regression surrogate on simulation-derived win bias so I can identify cards that overperform or underperform their printed stats.
- Used NetworkX board graph and layout analytics to study starting-zone balance, slot usage, frame centrality, and small board-value changes.
- Kept Streamlit as a local inspection demo over the same simulator, useful for changing seeds, strategies, and simulation counts while reviewing the results.
- Connected the outputs to a playable Expo / React Native prototype with bot policies and generated ML scoring artifacts, while keeping the physical board-game version as the next product direction.
Simulation and ML notes
- Used simulation as the measurement layer because deck order, player strategy, and board position create too many combinations for manual testing alone.
- Kept the ML layer small and interpretable: the ridge surrogate predicts simulation-derived win bias, and the Expo bot scorer ranks legal actions instead of inventing moves.
- Separated the digital prototype, balance pipeline, and physical manufacturing direction so each part can improve without overstating what is finished.
Current state
- The project currently includes the parsed card data set, Monte Carlo tournament engine, ridge-regression balance surrogate, board balance reports, local Streamlit inspection demo, and Expo prototype.
- The current board balance run improved overall balance from 0.5637 to 0.5688 and reduced the candidate zone gap from 0.3067 to 0.1484.
- The physical version is still in progress. I am using the data pipeline and app prototype to guide the next card, board, and manufacturing decisions.
Media
