Respond AI

Respond AI helps specialists guide long-distance carriers to their next pick-up and drop-off locations.

With an onboard LLM accessing data in nearly real-time, operators now begin with responses pre-populated. They perform a review check, and can focus on adding the human contact that drivers value for the long haul.

Shipping is decentralized, with many different companies running call centers to keep drivers up to date. It takes people time to query data tables and provide the most up to date information. Meanwhile, more messages from drivers keep coming in, leading to costly delays.

Every 30 minutes a fleet is delayed translates to $1.37 million in lost revenue.

Call centers route drivers dynamically to keep everything rolling. But it takes time for human operators to figure it out. What if an LLM could minimize delays?

In designing Respond AI, I created an agnostic solution that can integrate with existing data streams to pre-draft helpful responses. Making this human-in-the-loop keeps operators in the center, with the ability to make edits and add their human insight.

Doing so also provides a runway for operators to rate and annotate model outputs, improving the quality and accuracy of responses.

A central dashboard that helps operators get oriented for the day’s work

Each operator begins their day with a prioritized action list, real-time message and email queues, and a quality score tracking AI response accuracy. The command center surfaces what matters most so operators can focus on high-value conversations.

Measuring progress against daily and quarterly goals

The reporting dashboard gives team leads visibility into emails per day, response times, and quote volumes across teams. Operators set daily goals and track trends over time, creating a feedback loop that drives continuous improvement.

AI pre-drafts responses so operators can review, edit, and send

The messaging interface gives operators a queue of driver conversations on the left and a threaded view on the right. The LLM pre-drafts each response using real-time data — operators simply review, make edits if needed, and accept. This same pattern extends to mobile follow-ups, where drivers get quick, contextual answers on the road.

Templates that steer model outputs and ensure quality

To guide the AI’s responses, I designed a template system built around variables. Operators create reusable templates with dynamic fields — shipper, date, route, weather — that inform the mega-prompts and multi-shot prompting behind each response, ensuring consistency and accuracy.

Overwhelmingly positive response from active customers

Using click-through prototypes, several rounds of tests were run with active customers who use the current product. I wrote a conversation guide to structure the interviews, as well as set up a hypothesis board. I educated the team about research methods.

The feedback was overwhelmingly positive, so much so that the company president asked me to look at their other software applications and find opportunities.