AI and Investing: David Trainer’s Presentation at Harvard Business School

29 Apr 2026 · 1 min

Watch on YouTube

1:11

David Trainer, who's been working in machine learning since 2003, spent 20 years building technology to automate the reading of SEC filings. Companies file 200 to 500 page documents every quarter, and his team created explicit instructions to extract the critical data instead of hiring offshore workers to read them manually.

Chapters

I'm probably one of the sort of oldest school AI people you'll ever meet. I got my start really in machine learning back in 2003. And the problem we wanted to solve is that reading a 200 to 500 page 10-K or 10-Q was really just not practical anymore to do at any kind of scale for more than a handful of companies or even a single company, and certainly not every quarter for every company in a portfolio. And yet there's important Analytical information, critical oftentimes data in these 10-Ks. And we set about building that technology over 20 years ago, and we did it in really the old-fashioned way. We started gradually building up a library of very explicit instructions as to how these very nuanced and complicated disclosures should be treated. We did it all in a machine environment with experts. So you look at the traditional data providers, the legacy data providers, and they've usually solved the problem with multiple or thousands of people in an offshore location, sort of just throwing numbers of people at the problem. We decided to throw technology at it.