NORTH POLE HOTLINE

TYPE

Personal Project

ROLE

Conversational UX Design

TIMELINE

December 2025

TOOLS

Retell AI, GPT-4, ElevenLabs

A Santa voice agent for children to call that secretly encourages healthy habits through play.

+1 (341) 217-8981

Call Santa Bot now

OVERVIEW

The Concept

Santa Bot is a Santa voice agent for children to call. The agent mimics Santa's mannerisms and picks conversation topics relevant to Santa, Christmas, and the North Pole to keep children entertained.

The twist? The agent is secretly designed to encourage healthy habits through play. Santa Bot embeds conversations about "good deeds" that a child can do to stay on the "nice list" into its engaging chatter about Christmas. This motivates children to keep doing good deeds.

The Approach

I built Santa Bot using a combined LLM + prompt engineering approach through Retell AI Studio for conversational agent creation and GPT-4 for voice chat capabilities. I also leveraged ElevenLabs voice cloning to provide an affective, grandfatherly voice for my Santa bot.

Through rapid iteration and prototyping, I conducted dozens of experiments testing the naturalness, intuitiveness, and affective nature of the system across both "happy path" and adversarial scenarios.

The Origin

I made Santa Bot as a personal side quest to surprise my little cousins this Christmas.

They loved it.

DESIGN DETAILS

Intuitive Transitions

Rather than immediately accept a child's "No, I don't want a gift," or repeatedly press the same question, the bot applies balanced persistence — it makes a gentle second offer, then gracefully moves on if the child declines again.

Knowing When to End

The original baseline system didn't know when to end a call, endlessly keeping the conversation going even after accomplishing its goals. This led to user fatigue. I modified the conversation flow to add clear markers of what would signal an "end" to the call.

First Greeting

I intentionally defined the entry point as a callable phone number and configured the agent to greet the caller first (instead of the other way round) to mimic how human callers on "hotlines" typically function.

While an AI agent is "intelligent" on its own, imbuing it with the manners of a human is a complicated task requiring effort and thoughtfulness. This project taught me that iterative prompt engineering and testing is an effective way of improving conversational agent performance.