QUORUM
Completed
Built a Congressional vote prediction system achieving 95.7% accuracy on partisan legislation by training gradient boosting models on 1.6M historical votes across 31K+ bills with 75+ features per legislator.
PythonXGBoostFastAPIPostgreSQL
Architecture
flowchart TB
subgraph Data Pipeline
A[VoteView CSV] --> B[(PostgreSQL)]
C[Congress.gov API] --> B
B --> D[Feature Engineering]
D --> E[training_data View]
end
subgraph Model Training
E --> F[Time Split\nTrain: ≤115\nTest: ≥116]
F --> G[Preprocessing\nStandardScaler\nOneHotEncoder]
G --> H[GradientBoostingClassifier\n200 trees, depth 5]
H --> I[vote_predictor_v2.pkl]
end
subgraph API + Frontend
I --> J[FastAPI]
B --> J
J --> |/members\n/bills\n/predict/batch| K[Next.js War Room]
K --> L[435-Seat Chamber\nVisualization]
K --> M[3-Wave Vote\nSimulation]
end