← Back to selected projects

AI Product · Compass AI in product

Prompt Coach: designing the human side of AI

A 72-hour hackathon build that took a cross-functional team of six from concept to production-ready, won the #MoveFast Award at the 2023 Compass Hackathon, and shipped as part of Compass AI, where agents use it today.

72 hrs
Concept to production-ready
6
Cross-functional team, one sprint
#MoveFast
2023 Compass Hackathon Award
Compass AI chat interface showing topic options: Listing Description, Team Biography, Social Media Post, and Marketing Ideas

Compass AI today: Prompt Coach's topic-first approach, live in production.

The problem

Real estate agents needed personalized, platform-ready marketing copy, and generic AI output wasn't cutting it. Before Prompt Coach, agents either avoided AI entirely or pasted generic requests into Compass AI, our own AI embedded in the platform, and got copy that sounded nothing like them. The missing piece wasn't the AI, it was the UX layer that helped users give the AI what it needed.

Discovery: straight to the agents

One of our six was an Agent Experience Manager out of the Compass Malibu office. She connected us with agents who walked us through their frustrations firsthand: generic output, copy that didn't sound like them, prompts they didn't know how to write. Those conversations shaped the concept, and we tested the tool with the same agents before submitting.

Solution: the interface does the prompting

Prompt Coach guided agents through a structured workflow, auto-populating MLS listing data and capturing tone, audience, and brand voice upfront. By the time the AI generated copy, it sounded like the agent, not a template. The user never had to learn prompting, the interface did that work invisibly.

Diagram of the Prompt Coach flow: agent steps through choosing listing type, reviewing MLS data, adding a personal description, and defining voice and audience, while the system layer establishes context, auto-fills MLS data, captures agent voice, and assembles the prompt the AI model generates from

The interface does the prompting: a simple 4-step guided form for the agent, a fully structured, personalized prompt for the AI, invisibly assembled underneath.

Compass AI writing tool, Listing Description topic selected Compass AI writing tool, Team Biography topic selected Compass AI writing tool, Social Media Post topic selected Compass AI writing tool, Marketing Ideas topic selected

Listing Description — Prompt Coach, live in production as Compass AI: four guided topics, each pre-loaded with the context the model needs to sound like the agent, not a template.

See it in action

The hackathon prototype: the guided flow agents step through, from choosing a listing type to landing on personalized, on-voice marketing copy.

Interactive, click through the prototype above.

The team: 72 hours, six people, one shipped feature

I led UX and UI across a cross-functional team of six, including the Agent Experience Manager and front-end and back-end engineers, taking the concept to production-ready in 72 hours.

Outcome

Prompt Coach won the #MoveFast Award at the 2023 Compass Hackathon and shipped to production, where agents use it as part of Compass AI. The hackathon proved the idea viable and shippable, the product team carried it forward from there.

#MoveFast
Award, 2023 Compass Hackathon
Shipped
Live in production as part of Compass AI
2023 Compass Hackathon #MoveFast Award slide for Compass Prompt Coach, with the team: Amanda Cathey, Kfir Ankori, Aditya Muthyala, Yang Liu, Pradeep Konda, and Rachel Russell-Clark

The #MoveFast Award, 2023 Compass Hackathon, with the six-person team behind Prompt Coach.

What I'd take from this

Prompt Coach taught me where the real design problem in AI lives. It isn't making the model smarter; it's helping humans express intent. Agents knew exactly what they wanted, they just had no way to say it to a machine. That translation is a UX problem, and it's the lens I've brought to every AI project since.

What I'd do differently: stay closer to the feature after handoff. The 72-hour sprint was the headline, but the learning I gave up was in the long tail of how agents actually used it.

Role

Product design lead

Industry

Real estate tech

Timeline

72 Hours