<?xml version="1.0" encoding="UTF-8"?><oembed><type>video</type><version>1.0</version><html>&lt;iframe src=&quot;https://www.loom.com/embed/a2cd9192359840cab5274ccba399bd87&quot; frameborder=&quot;0&quot; width=&quot;1404&quot; height=&quot;1053&quot; webkitallowfullscreen mozallowfullscreen allowfullscreen&gt;&lt;/iframe&gt;</html><height>1053</height><width>1404</width><provider_name>Loom</provider_name><provider_url>https://www.loom.com</provider_url><thumbnail_height>1053</thumbnail_height><thumbnail_width>1404</thumbnail_width><thumbnail_url>https://cdn.loom.com/sessions/thumbnails/a2cd9192359840cab5274ccba399bd87-42708e8f68b7c60d.gif</thumbnail_url><duration>107.007</duration><title>Maitai Autocorrections Demo</title><description>In this video, I demonstrate how Maitai simplifies building reliable AI applications by detecting and autocorrecting faulty outputs from large language models in real time. Using our demo site, I showcase selecting a model, such as gp4-4o, and enabling autocorrections. Through a scenario where I try to coax the ordering agent to give a discount against its instructions, you&apos;ll see how Maitai corrects the model&apos;s output to maintain accuracy.</description></oembed>