Conversational AI Before Foundation Models

Paradox builds AI that automates recruiting and hiring for global enterprises like McDonald's, Unilever, and FedEx. I joined in 2018 as the company's first AI Product Manager, before foundation models were a household name, when much of the work was defining how a conversational assistant should behave, scale, and earn trust in high-volume, compliance-heavy workflows. Joining this early meant building without a roadmap. Many of the questions teams ask today about conversational AI, product adoption, and human-AI interaction were still being figured out in real time. That experience shaped how I approach every new technology: prototype early, learn quickly, and stay close to real users.

As direct product owner, my flagship was the AI Content Management System. Customizing the assistant's responses used to take clients 6+ months of internal Paradox work, with no client-facing interface, no rule-based targeting, and static content that was losing engagement. I analyzed product and conversation data, interviewed Fortune 500 clients, ran design-thinking sessions with design and engineering, and wrote the PRD and wireframes. We shipped a clean interface where clients could configure responses, build rule-based audiences (geographic, status-based), add dynamic media like video, images, and PDFs per topic, and see analytics on what was performing. As the first AI PM I also spun up a lot of Paradox's 0→1 MVPs: the conversation builder, ratings and feedback (including emoji reactions over SMS and WhatsApp), multilingual support, and the dynamic career sites that became the award-winning Experience Assistant. I got each one out and in front of customers, then handed it to a dedicated PM as we scaled. I also worked on the MVP and global go-live of McHire, McDonald's hiring app for candidates and store managers, where applying and onboarding that used to take months now takes hours to days.