As Connect’s AI-powered chatbot automated institutional inquiries, human input became essential to growing the system’s knowledge base. This project explores how over-automation can remove context and authority—and how redesigning for human judgment, timing, and agency ultimately led to service improvements.

As Connect expanded to multilingual institutions, language emerged as more than a translation problem. This project reframes language as infrastructure—coordinating AI responses, human intervention, UI systems, and client-authored content to preserve meaning and consistency across languages.

Early versions of Connect relied on human support to handle errors, creating a warm first impression but limiting long-term system learning. This project examines how feedback design shapes participation—showing how reducing emotional and cognitive burden helped users engage more openly and enabled sustainable, reliable feedback that improved system accuracy.