From household questions to answers to actions— AI handles the details so families don’t have to.

Families can submit questions about daily tasks or activities, and AI generates tailored suggestions and actionable results

Ai Agent always asks follow up actions

Would you like me to call these options to get more informations?

Do you want me to help book an appointment for you now, or check anything else first?

Want me to add them to your calendar?

AI offers to take real-world actions: make calls, ask additional questions, or schedule on the user’s behalf.

All requests are tracked in home page, showing their status: researching, scheduling, or waiting for user input if additional information is needed to complete a task.

Listen to the AI call »

Asking Questions, Receiving Curated Results

From Answers to Action

Tracking Requests in One Place

AI That Acts on Your Behalf

We used Blend AI (voice-AI platforms) to explore how AI could handle real-world follow-ups that often block families from completing tasks. By allowing the AI to place phone calls, gather availability, and return summarized results, the experience reduced interruptions and enabled families to make informed decisions quickly—without losing control over the final action.

Making AI-initiated calls feel natural required careful prompt design. We shaped how the agent spoke, listened, and paused—teaching it to recognize live conversations, voicemail, and automated systems. By designing for these real-world nuances, the calls felt smoother and more respectful, creating a better experience for both users and businesses.

Designing Conversations, Not Just Calls

Designing the Nuances of Conversation

Challenge 1

Designing Beyond the Happy Path

Most real-world tasks don’t resolve perfectly on the first try.

The ideal flow—ask, choose, book—was only part of the experience.
We needed to design for moments when AI couldn’t complete a task smoothly: unanswered calls, closed businesses, long queues, or results that didn’t meet expectations.

The challenge was defining how the system should recover—when to update users, what alternatives to suggest, and how to keep tasks moving forward without causing confusion or frustration.

Challenge 2

AI Accuracy & Context Awareness

If AI responses aren’t precise, even the most thoughtful design can fail.

Developers used an AI testing tool to validate responses before integrating them into the product. We found that results were highly sensitive to how questions were phrased—missing or unclear details like location or dates often led to inaccurate or outdated results. Improving context awareness was critical, as unreliable AI output would quickly break user trust.

Example prompt included a target date, but the AI surfaced outdated festival information.

unanswered calls

closed businesses

long queues

Unexpected results

Call immediately drops

Carrier rejects the call

Spam detection

Interactive Voice Response rejection

Human in the Loop

Automated phone calls can fail in many unpredictable ways. Instead of canceling the request or leaving users without an outcome, we designed an escalation path as part of the system. After multiple failed attempts, the task would be handed off to a human in the loop, ensuring the work can still be completed and the user isn’t left stuck in an automation loop.

When Automation Stops, a Human Steps In

Good Ideas Don’t Always Survive Reality

What This Project Taught Me

This project reminded me that strong design ideas still need to survive technical constraints. With a short timeline and a heavy focus on AI and backend performance, many design decisions depended on whether the system could reliably generate accurate results. We had to let go of certain UI ideas to prioritize what mattered most: whether the AI could actually deliver value to users

Designing Trust Into the Experience

Beyond scope trade-offs, the bigger challenge was designing around AI’s limitations. The experience depended on whether AI could correctly interpret intent, communicate with real people, and return accurate results. When AI behavior was inconsistent, even small errors risked breaking user trust—making reliability as much a design concern as a technical one.

Designing for What AI Can Actually Do

Beyond scope trade-offs, the bigger challenge was designing around AI’s limitations. The experience depended on whether AI could correctly interpret intent, communicate with real people, and return accurate results. When AI behavior was inconsistent, even small errors risked breaking user trust—making reliability as much a design concern as a technical one.