Inspiration
Professional sports front-offices routinely destroy their financial flexibility by overpaying for big-name superstars based on media hype, past reputations, and narrative bias. We built this project to remove the human emotional element from roster construction. By completely masking player names and compensation tiers during the scouting phase, we empower front offices to locate undervalued talent based purely on objective, un-hyped on-court production.
What it does
Moneyball 2.0 acts as an automated "clone" search platform. Users select a high-priced superstar and set a strict financial ceiling slider. The application immediately processes the target player's multi-dimensional statistics against the rest of the league database. Using distance calculations, it filters out the high-priced star and highlights the top five cost-effective players whose statistical profiles mirror the superstar, showing GMs exactly how to replicate elite production at a fraction of the price.
How we built it
Our system architecture relies on an entirely decoupled development pipeline. The data layer features an automated data-cleaning environment that normalizes column formats, corrects trailing string whitespaces, drops incomplete data profiles, and outputs a strict numeric matrix. The front-end interface is entirely engineered using a modern python-based dashboard ecosystem, deploying interactive scatter plots to track efficient value frontiers and customized multi-axis polar grids for side-by-side performance comparisons.
Challenges we ran into
Our primary hurdle was scale bias during distance calculations. Standard volume stats like raw points completely overwhelmed crucial ratio metrics like field goal percentages, which skewed the early matching results. We solved this data-drift anomaly by implementing a dynamic scaling feature that forces all parameters into an identical index scale before calculating vectors. This mathematics tweak was fully validated when our live test run successfully mapped elite budget alternatives within crisp, non-overlapping visualization charts.
Accomplishments that we're proud of
Our team engineered a fully decoupled, asynchronous development pipeline where Track A built a resilient Python backend while Track B simultaneously structured the Streamlit user interface using dummy data contracts. To ensure accurate player matching, we integrated a MinMaxScaler normalization pipeline to prevent high-volume metrics (like points per game) from statistically overpowering crucial efficiency percentages. This similarity architecture successfully passed validation by correctly cloning elite superstar data profiles—like Stephen Curry—to highly cost-effective alternatives such as Ja Morant and Anthony Edwards.
What we learned
We discovered the incredible predictive strength of spatial similarity tracking by plotting normalized multi-axis player profiles within an n-dimensional geometric coordinate zone to calculate spatial distance. This vector-based approach proved that stripping away brand hype, player names, and media bias forces true, data-driven blind scouting objectivity that can help small-market sports franchises bypass major financial restrictions. Furthermore, we mastered rapid dashboard deployment using Streamlit, learning how easily a localized machine-learning project can be wrapped into a professional frontend user interface in under 50 lines of clean code.
What's next for Moneyball 2.0: The "Blind Scout" Value Optimizer
Our next operational upgrade is adding a position filtering module into the backend matching engine so General Managers can narrow down their alternative searches explicitly to replacement Guards, Forwards, or Centers. We also plan to introduce interactive sidebar sliders to let users dynamically adjust metric weighting weights, allowing a team analyst to double the priority of offensive statistics over defensive rebounds depending on their roster strategy. Finally, we intend to scale our data ingestion architecture to handle completely different sports fields, including multi-variate European soccer transfer market data and real-time Formula 1 driver telemetry feeds.
Log in or sign up for Devpost to join the conversation.