A year ago, U.S. President Donald Trump shut down public access to the Development Experience Clearinghouse, a $30 billion database holding 60 years' worth of institutional knowledge from more than 150,000 projects administered by the U.S. Agency for International Development. But before the closure, former USAID employee and artificial intelligence scientist Lindsey Moore used a large language model (LLM) to read all of the information in this database — rescuing critical lessons on development, environmental, economic and social projects in countries across the globe, all documented by USAID.

The data also included information on conservation projects. Many of the challenges presented in these projects repeated over the years, but the lessons were rarely retained — something Moore's tech startup, DevelopMetrics, hopes to change.

Moore joins this week's podcast to explain what those lessons are and what conservationists can learn from them. DevelopMetrics deploys an AI model capable of understanding not just the information from USAID's database, but also other public databases that could be at risk of deletion or being lost to time.

Please take a minute to let us know what you think of our podcast, here.

Mike DiGirolamo is the host & producer for the Mongabay Newscast based in Sydney. Find him on LinkedIn and Bluesky.

Banner image: Mangroves on Vanua Levu Island, Fiji. Image by Rhett A. Butler/Mongabay.

——

Timecodes

(00:00) Lindsey's background with USAID

(04:14) How to analyze 60 years of data

(11:07) Uncovering hidden lessons

(14:24) 1. Bring delivery closer to households

(16:43) 2. Practice changes practice

(19:19) 3. Design for scale, not for pilots

(24:08) 4. Co-creation beats consultation

(26:30) 5. Strengthen the middle layer

(30:56) Who DevelopMetrics works with and how they are funded

(32:58) Energy and water costs of LLMs