Human-in-the-loop

Goal

As the team continued to build the foundations of the systems, we found that we’ll need human-in-the-loop interactions for users to approve the tasks that the AI was performing. We started with this exploration and iterated this design when the team worked on agent transparency streams for the users

Building the flows

I collaborated with the Engineer and Product leads to highlight when, where and what will be triggered when a customer interacts with the AI chatbot. Creating the initial flows was important for:

  • Maintaining the experience layer of human-in-the-loop

  • Understand the different states the AI will experience

  • Influence the experience how the AI will response back to the user

  • Identify gaps that the AI might respond back to the user

Identify states & tools

During the design phase I worked closely with the engineer team to identify the tools that the customer can prompt for the AI. For MVP, we are releasing with the digital interconnection portfolio. The main states were identified and I need to create the different states for the team to implement and identify gaps.

Final Designs - Customer Portal, Slack & Teams

I had to align on a interaction that had continuity across 3 different platforms. This was challenging because of the SDK constraints I had with the other platforms. I collaborated closely with the another engineer team to maintain the integrity of the experience and made sure that the AI was responding correctly to the customer with the human-in-the-loop is triggered.

Challenge

The AI produced confusing, varied outputs. I worked closely with the engineering team, the draft of MCPs, and the APIs for this experience. I mapped the basic flow and iterated quickly.