© Illustration by Jean-François Desserre
Jason Langsner is Group Product Manager, Data + AI, at The Washington Post. In this article he shares how they built The Post’s in-house generative AI chatbot tool that uses RAG to answer user questions.
Key takeaways to copy & share
> The Washington Post built Ask The Post AI, a chatbot answering reader questions using retrieval augmented generation over its own reporting from the past eight years, with links to every source article > It started as one desk rather than a platform. Climate Answers launched with the Climate Desk, drawing only on Post climate coverage, before the approach was scaled site-wide using Meta's open-source Llama models > The hard problem isn't answering. It is teaching the model to say when The Post hasn't covered something well enough to give a solid answer, which is the work Jason Langsner singles out > It came out of a five-day AI hackathon in 2022, run by engineering and product alongside the newsroom. That became a permanent R&D lab, and The Post now has a Chief AI Officer > Internal first, external second. The lab builds tooling for Post staff as its own customers before features reach subscribers, so the newsroom stress-tests them before readers do
The Washington Post may be best known for our world-class and award-winning reporting, but we also pride ourselves on being industry leaders in technology and innovation. In addition to producing journalism, we also develop products and solutions in house to improve our users’ experience, including through our expanding AI Pod Research & Development (R&D) lab and newly launched WP Incubator.
