HeroCall

The problem

Every parent knows the daily negotiations: bedtime, teeth, homework, toys off the floor. Kids tune out the same request from the same voice, yet they light up for a hero. We asked whether a hero could make that call for real — a live, two-way conversation that encourages a child instead of lecturing, and is safe enough that a parent hands over the phone without a second thought.

What we built

HeroCall lets a parent schedule a real-time voice call from a superhero to their child. The hero talks with the child about the mission, gets a promise, celebrates when it is kept, and remembers them the next time it calls. The cast includes 20 original HeroCall heroes; five fairy-tale and legend heroes — Santa Claus, the Snow Queen, the Monkey King, the Tooth Fairy, and Starsong; and custom heroes each family creates in Hero Studio.

• Scheduled hero calls for bedtime, teeth, homework, tidy-up, and more
• Missions built on positive-discipline techniques, with stars and streaks
• A small memory of each child that parents can view and edit
• Every hero is an original or from the public domain
• Available in 12 languages

How it is built

Each call runs a streaming voice pipeline: Whisper turns the child's speech into text, a hero agent built on Google ADK decides what to say, and the hero's own acted text-to-speech voice answers, streamed sentence by sentence so the conversation feels live. After the call, a Memory Keeper agent updates the child's memory — hobbies, routines, wins. Safety is designed in from the start: a parent consent gate, multilingual child-safety detection, a check on every hero sentence before it is spoken, and no audio stored.

• Whisper speech-to-text → Google ADK hero agent → per-hero acted TTS
• Sentence-by-sentence streaming for a live conversation
• Memory Keeper agent maintaining a parent-editable memory
• Consent gate, multilingual safety detection, no audio stored
• Flutter mobile app

The result

HeroCall is in development. Its voice pipeline, missions, memory, and safety layer all serve one rule: a child should only ever hear what a careful parent would approve. It is the kind of real-time, multi-agent voice product ISEMI builds — speech streamed in and out, agents that remember, and safety checks that run before every sentence reaches the listener.