← All work

South Park Commons · Independent · 2025

Prototyping human-AI products beyond the chat box

Three working bets on AI characters, task automation, and cognitive support — each built to expose a different trust problem.

AIPrototypingHuman-computer interaction
RoleProduct designer and builder
Timeline2025
TeamIndependent builds with technical collaborators and AI tooling

Measured impact

01

Generated hundreds of culturally grounded character configurations

02

Built an Android launcher in Kotlin around instruct-model task execution

03

Prototyped multimodal cognitive training for senior citizens

Three products, one question

During my time as a South Park Commons explorer, I investigated what changes when AI becomes part of a product’s behaviour rather than a feature added to its surface.

Culturally grounded characters

I built an agentic system that could create hundreds of AI character configurations for dating and social interaction. The hard part was not generating biographies; it was preserving cultural specificity, behavioural consistency, progression, and safety across repeated conversations.

An intent-led Android launcher

In Kotlin, I prototyped a launcher that used instruct models to plan and automate multi-step phone tasks. Working code exposed the gaps between interpreting an intent and safely taking action: confidence, permissions, previews, reversibility, and fallback interfaces.

Cognitive training and support

I also prototyped a conversational application for senior citizens, combining cognitive exercises with a supportive interaction layer. Trust was not a tone-of-voice problem. It depended on clarity, pace, memory, accessibility, and showing the user what the system believed was happening.

Across all three, the most useful design material was the failure state. AI products become legible when uncertainty and control are treated as first-class interface elements.

Continue exploring

Different medium.
Same builder instinct.

More work ↗Enter the lab ↗