Built a proof-of-concept AI sales agent for a CRE top-of-funnel workflow (lead intake → qualification → handoff). The point wasn't to "ship AI" just to say we did. It was to sanity-check what would actually be useful in a loan officer's day, design a workflow that wouldn't feel annoying or extra, and pressure-test prompts + guardrails before anyone invested real engineering time.

I treated it like a practical product experiment: map the real-world flow, figure out where an agent could genuinely remove friction (without making things feel robotic), and define what "good" looks like in the handoff. The outcome was a clear, testable flow and a set of prompts designed for real usage, not a demo.

With an applied math background, I'm comfortable rolling up my sleeves and learning new tools quickly. I started this project before agent-building tools became mainstream, so I prototyped in Relevance AI. If I were to build it out fully today, I'd build it from scratch rather than rely on a platform that limits customization.