Built for Business and Functional Heads at Stanza Living. It brought natural-language access to operational data into a React application, backed by Redshift and a custom agent harness. I led the work with two developers, collaborating with the Data Science team.
The obvious automation was not the whole problem.
The product needed to support more than a single question and answer. Leaders needed to explore operational information in context and keep track of updates from ground teams. The challenge was designing a system that could support those connected needs.
The interface made the system approachable. The harness made memory and context part of how it worked. That placed much of the product effort beneath the chat experience, in deciding what information the agent needed and how to carry it through the workflow.
Build the operating loop, not only the intelligent step.
Anchor the experience in leadership workflows. Design for Business and Functional Heads asking operational questions and coordinating follow-up.
Build the agent harness. Work with two developers on the infrastructure around the model, including memory and context engineering.
Connect the data capability. Collaborate with Data Science on the NL2SQL capability and its connection to Redshift.
Support operational communication. Include support for organizing updates and communication with ground teams.
Shipped a React-based operational copilot for Business and Functional Heads, combining natural-language access to Redshift with a custom agent harness built around memory and context, while helping organize ground-team updates and communication.
What stayed after shipping.
The most substantial work sat beneath the interface. Building the harness around memory and context was central to making the agent useful within an operational workflow.