FrizzleNet · Kaggle Hackathon 2026
Designing for Parent-Child Co-Regulation & Restorative Play with Agentic AI.
A systems design approach to conquering "Manager Brain" burnout through AI-guided physical play.
- Role
- Lead AI Experience Designer & Design Technologist
- Team
- 2 People (UX + Engineering)
- Stack
- Google Antigravity (ADK), Custom Prompt Frameworks, Google Maps Geofencing APIs
- Deliverable
- Interactive Place-Based Learning Agent
01 · Overview
What FrizzleNet is
FrizzleNet is an agentic AI companion built on Google's Agent Development Kit that turns "what should we do today?" into a five-second answer — matched to the family's location, the child's developmental stage, and whatever supplies are already in the recycling bin. Built in two weeks for the Kaggle Vibe Coding Intensive, it explores what happens when AI is designed to reduce screen time instead of capture it.
02 · Challenge
Deconstructing 'Manager Brain'
Modern parents don't need more ideas — they need less overhead. Toggle between the typical planning loop and the FrizzleNet flow to see the cognitive load shift in real time.
Cognitive Load
Manager Brain
03 · Systems Thinking
Co-Regulation Interaction Model
FrizzleNet reframes the parent-child interaction so both people leave the activity regulated. The parent stops directing traffic; the child steps into agency.
Role shifts from Active Cruise Director to Calm Anchor / Passive Supervisor.
Cognitive Load
10%
Active Engagement
Low
Emotional Support
High
04 · Live Sandbox
Vibe-check the agent
A miniature previewer of the FrizzleNet agent. Change the location, mode and age — the activity card responds like the real ADK output.
Cardboard Castle Builder
Materials
- Recycle bin items
- Masking tape
- Markers
Parent Guide
"What do you think lives inside this castle?"
Parent-Child Conversation Starter
- → Who built this castle and why?
- → What's the secret room nobody knows about?
05 · Narrative
Systems Thinking: Designing Beyond the Screen
FrizzleNet started from a hypothesis: most "family AI" products optimize for engagement metrics that quietly compete with the relationship they're supposed to serve. The design brief inverted that — build an agent whose north-star metric is successful disconnection. The physical-digital co-regulation framework maps every AI response to a concrete, offline moment between an adult and a child, with the digital layer disappearing as soon as it hands off.
This meant designing the prompt and response contracts around a three-actor system — parent, child, environment — rather than a single user. Read the original writeup: FrizzleNet Kaggle project writeup .
06 · Stack
Information Architecture & Tech Stack
Google Antigravity + ADK
Multi-step agent workflows with state machines for the "suggest → adapt → hand-off" loop. Dynamic prompt engineering structures encode developmental stage, materials, weather and geo-context as first-class inputs.
Google Maps Geofencing
Latitude/longitude, radius and landmark type feed a geofencing layer that filters for free, public, family-safe destinations before the LLM ever sees a candidate.
Prompt Frameworks
A custom "Anchor / Autonomy / Artifact" template — every response must produce a parent anchor line, a child autonomy task, and a physical artifact instruction.
HITL Feedback Loops
Lightweight thumbs-up / rewrite affordances feed back into prompt tuning so suggestions get more locally relevant over time.
07 · Reflection
Key Outcomes & Reflections
The most interesting result wasn't a metric — it was a reframe. FrizzleNet works when the parent stops feeling like a project manager and starts feeling like a companion. That's the interaction the agent is really designing for: a nervous system that softens.
For me it validated a design principle I keep returning to: AI is at its best when it helps people put it down. The agent's job is to load up the parent with just enough structure to be calm, then get out of the way while the actual play — the actual relationship — happens offline.