Hapi

Inspiration

My inspiration for this project came from my experience working in Child Protective Services. During that time, I worked with a lot of families that needed all kinds of assistance. Many of the children I encountered lacked a consistent, safe emotional outlet — someone or something they could talk to without fear of judgment. I saw firsthand how early disclosure and gentle communication could change outcomes, but the systems in place often made that difficult. Hapi is my attempt to bridge that gap: a quiet, trustworthy presence for children in vulnerable situations, designed to support the work of caregivers and caseworkers rather than replace it.

What it does

Hapi is a voice-first AI companion app for children ages 4–6, built in the Jac/Jaseci framework with a React frontend. It presents as a friendly minimal face avatar that children can talk to via tap-to-talk voice input or typed text, and responds with spoken and displayed replies tailored to young children. Underneath the friendly surface, Hapi runs a hidden multi-stage safeguarding pipeline that silently analyzes every message for concern categories like physical harm, abuse, self-harm, bullying, and unsafe environments. Low-confidence concerns are noted, medium-confidence ones trigger a gentle follow-up question, and high-confidence disclosures generate an immediate structured report — immutable, with a full audit trail — delivered as a mock CPS-style email to a designated caseworker inbox. The app also maintains a selective memory of benign child preferences, while blocking any sensitive disclosures from being stored in regular memory. It supports four interaction modes (companion conversation, feelings check-in, tiny story generation, and a fun fact mode), all governed by the same safety pipeline, with Google Cloud Speech handling voice input and output and Claude Codes Haiku powering Hapi's logic.

How we built it

It was built with Jac, the programming language, and the Jaseci framework. Utilizing codex and Claude Code for AI assisted development.

Challenges we ran into

• Feature creep
• Some setup issues -Conflicting program logic issues.

Accomplishments that we're proud of

Seeing the MVP realized.

What we learned

-Jac -Jaseci -Solo development -API keys -Workflow Optimization

What's next for Hapi

Further development into a full fledged product.

Built With

• anthropic-claude-code-haiku-api
• google-speech-to-text-api
• google-text-to-speech-api
• jac
• jaseci