Haystacker
An AI-powered forensic journalism tool born in a hackathon.
Modern journalists have unprecedented access to public data. Events are recorded from countless angles and shared online for everyone to see. Political campaigns across the world upload advertisements to online platforms as soon as they might otherwise have only aired on local television.
Stories are hidden within and across all of this media, but just as newsroom access has never been greater, resources have never been more limited. How might technology empower forensic journalists to find the proverbial "needle in the haystack?"
This was the question posed by my colleagues in the The Washington Post newsroom organization. During a hackathon I led design for the team that proposed and ultimately built a solution: a new AI-powered reporting tool called Haystacker.
By using AI to label large volumes of media, Haystacker saves countless hours of painstaking labor that would otherwise be spent manually combing through videos one by one. After demonstrating a prototype solution during the hackathon our team was granted the resources to design, develop, and deploy a shipping product.
I defined the vision for a maximally useful initial release while laying the groundwork for future features. The first iteration was asset-focused, with AI processing videos and labeling them for analysis by the newsroom. By using machine learning for large-scale pattern recognition, Haystacker creates space for journalists to do what they do best: in-depth human storytelling.
Label view is oriented around filtering label instances across an entire project.
Soon after Haystacker's release to the newsroom, the Post published the first story reported using the tool. 745 campaign advertisements were processed by Haystacker and then analyzed by forensic journalists. I'm proud to have my name included alongside the storytelling that Haystacker makes possible.
The initial release was not the end of my of my contributions. The second major version focuses on labels, allowing journalists to find and filter label instances across an entire project instead of just within one asset at a time. This empowers the newsroom to accomplish even more using the original design foundation.
