<?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/198e9c8135fa4581abeb6d1b4208293a&quot; frameborder=&quot;0&quot; width=&quot;1866&quot; height=&quot;1399&quot; webkitallowfullscreen mozallowfullscreen allowfullscreen&gt;&lt;/iframe&gt;</html><height>1399</height><width>1866</width><provider_name>Loom</provider_name><provider_url>https://www.loom.com</provider_url><thumbnail_height>1399</thumbnail_height><thumbnail_width>1866</thumbnail_width><thumbnail_url>https://cdn.loom.com/sessions/thumbnails/198e9c8135fa4581abeb6d1b4208293a-f4c42aa0b515b6c2.gif</thumbnail_url><duration>299.968</duration><title>SenPilot Technical Challenge Submission — Muneeb Hassan</title><description>This Loom demos Vellum, an asynchronous event driven agent that automates downloading regulatory filing documents from a stateful web portal via email. The demo shows a user sending a Gmail message with a mandatory Vellum tag and including a matter number and document types, which triggers a FastAPI webhook and queues the job without IMAP polling. Logs confirm intent parsing using GPT 5.4 mini, loading matter M12383, spinning up Playwright instances to download files, packaging them into a zip, and emailing the user an HTML and plain text payload with a download button. The author also explains challenges with the UARP dashboard selectors due to dynamic IDs and notes plans to add vector database ingestion and improve multi tasking.</description></oembed>