Short answer: Eldar Sadikov says Field Materials has 20-plus agents that read documents like switchgear quotes and trucking tickets. As foundation models improve, its ability to read and verify documents improves too, plus its own fine tuning.
Watch this answer from Episode 33 (starts at 12:44) · Episode 33 notes and transcript
From Episode 33: AI Is Transforming Construction — We’re Entering a Golden Era, starting at 12:44.
Jordan Metzner [12:44]: Oh, that's awesome. And do you think can we just jump in a little bit to kind of the impact some of these frontier models have had? I mean, have the impact of kind of the new Opus models or the new Claude Code models been step function improvements in what you've been able to develop, not only I guess from your product perspective, but you know, internally for software development and things like that as well.
Eldar Sadikov [13:04]: Yeah. So if you can see on the screen right now, like you have all these different agents that we build in, so we have dedicated agents for different document types. So you have you know, there are different quote documents whether it's switchgear quotes, HVAC quotes, you know there's different documents like trucking tickets or asphalt tickets. Underneath it all, we're basically building a dedicated agent that can understand and read those documents and do the data entry for your understand what is what in that document to automate the verification and matching. Because we're using these foundational models, when these foundational models get better every single quarter, every single month, our fundamental ability to read, interpret, and verify the documents is also going up. So and that's that's the amazing impact of all these upgrades that we're seeing in the industry that you know, you sort of like, your models could get better overnight literally because simply somebody released a new version and it works so much better. So and that that's sort of one types of agent that, and now we have 20 plus agents that sort of read different types of documents. And we're gonna be introducing other agents very soon beyond just extraction and reading of the documents, but automating a lot of the steps itself in the ERP or sort of procurement steps that that need to take place. And because we're aligned on the foundational models underneath, obviously our models will get better in that end. And we also ourselves doing a lot of work just constantly refining the models themselves and training those models on thousands of documents that we have, sort of fine tuning them to operate better and better, and making a lot of enhancements to make them smarter.
