24cast.org

Overview

24cast.org is a sophisticated election forecasting platform that uses machine learning to predict election outcomes down to the margin. As Core Web Operations Lead from September 2023 to November 2024, I helped build this innovative platform that became a trusted source for election predictions during the 2024 cycle.

Our final prediction gave Kamala Harris a 70% chance of winning the presidency, based on 100,000 simulations using a machine learning model trained on data spanning multiple decades. The platform provided daily updates until freezing predictions at midnight on November 5th.

My Role

As Core Web Operations Lead, I was responsible for:

• Building the technical infrastructure using Next.js and AWS
• Implementing CI/CD pipelines that reduced release cycles from weekly to daily
• Leading web operations strategy from product roadmap to market launch
• Collaborating with a cross-functional team of 5 members
• Developing the front-end interface for data visualization
• Partnering with data scientists to surface journalist-friendly insights

Technical Innovation

24cast.org pioneered the use of modern machine learning techniques in election forecasting:

• Machine Learning Model: LightGBM (tree-based model) trained on elections from 2002-2022
• Polling Methodology: Innovative bias-adjusted inverse-variance weighting for maximum accuracy
• Data Sources: Integrated data from FiveThirtyEight, Cook Political Report, Sabato's Crystal Ball, and FRED
• Interpretability: Used SHAP (Shapley Additive Explanations) to make ML predictions understandable
• Daily Updates: Automated pipeline using GitHub Actions and AWS with DynamoDB

Key Results

Technical Architecture

The platform was built with a modern, scalable architecture:

• Frontend: Next.js with TypeScript for type-safe, server-side rendered pages
• Backend: AWS Lambda functions for serverless compute
• Database: DynamoDB for scalable NoSQL data storage
• ML Pipeline: R and Python for data cleaning, analysis, and model training
• Infrastructure: AWS with automated CI/CD pipelines via GitHub Actions
• Visualization: Custom interactive maps and charts for election results

Election Night

We hosted a major election night event at Brown's Salomon Center, featuring:

• Live broadcast with special guests and student/faculty commentary
• Real-time political analysis using our custom Election Portal
• Live race calls by our Decision Desk team
• Co-sponsorship from Brown Political Union, Watson Institute, and other campus organizations

Impact & Recognition

24cast.org represented a significant advancement in election forecasting methodology, being one of the first platforms to fully leverage modern machine learning techniques for political predictions. The project demonstrated how technical innovation can make complex data more accessible and understandable to the public.

Through this project, I gained invaluable experience in building high-stakes, public-facing applications that require both technical excellence and clear communication of complex information. The success of 24cast.org showcased my ability to deliver production-ready systems under tight deadlines while collaborating with diverse teams.

Press Coverage

• Get to know the election-result prediction model developed by Brown students

  The Brown Daily Herald

• New Election Prediction Model by Students at Brown University Outperforms Industry Leaders

  PR Newswire

• By the numbers: Bethesda student creates online election predictor

  Bethesda Magazine

• AI is Being Used To Predict Election Outcome With Mixed Results

  CCN.com