My Role

• Position: Staff Software Engineer
• Responsibilities: Team Lead, Conversational UX Design, Front-End & Back-End Development
• Collaborators: Hackathon Team, AI Specialists, Newsroom Leaders, Product Team

Summary

As part of The Washington Post’s AI/ML hackathon, I led a team to build an AI-powered elections chatbot. The bot could answer natural language questions about any race, explain our predictions, and link users directly to the appropriate election results page. It was built to help readers make sense of complex visualizations and live data across hundreds of races.

Problem

Election night presents a firehose of information, hundreds of races, dynamic visualizations, and predictive models that can be difficult for readers to interpret. We needed a way for users to ask simple questions like “Who’s winning in Georgia’s Senate race?” or “Why did we call this race?” and get clear, fast answers with links to live results.

Goals

• Help readers navigate complex election data through natural language
• Explain predictive model outputs clearly and accurately
• Support newsroom confidence with guardrails against AI hallucinations
• Build a prototype good enough to take toward production

Process

I formed and led a cross-functional hackathon team. I designed the conversational flow and technical architecture, combining retrieval-based responses with strong fallbacks to ensure accuracy. I also built the UI, integrated it with live election data sources, and demoed the chatbot to senior leadership. After winning the hackathon, I worked across departments to start the process of bringing it to production.

Design Decisions

• Used a retrieval-augmented approach with race-specific data
• Designed fallback responses when data wasn’t available or uncertain
• Built guardrails to prevent model hallucinations in high-stakes situations
• Created a conversational UX optimized for clarity and trust

Challenges

The biggest challenge was accuracy. Elections are high-stakes, and any incorrect or misleading answer could damage trust. We developed strict validation layers, fallbacks, and data checks to make sure the newsroom could rely on every response. Getting internal buy-in also required close collaboration with product, editorial, and AI safety specialists.

Impact

• Won The Washington Post’s AI/ML hackathon
• Selected for production investment by newsroom leadership
• Created a new model for how AI can support live journalism
• Planned for upcoming deployment to live election pages

Reflection

This project was a career highlight, leading a team from idea to demo, winning a hackathon, and then turning that prototype into a real product. It deepened my passion for combining AI with journalism, and showed me what’s possible when you bring together creativity, trust, and technical precision.