This is a submission for the [Hacktoberfest Weekend Challenge: Build for a Friend]
##What I Built
I built Unbox, a voice-first productivity tool for a friend who constantly has ideas, tasks, meetings, reminders, and random thoughts bouncing around in their head — but doesn't always have the time or energy to organize them.
The idea was simple:
Don't make people organize their thoughts while they're having them.
With Unbox, you can simply open it and talk.
“Tomorrow I need to finish my DBMS assignment. Remind me to call Mom at 7. I also have a meeting with Rahul tomorrow.”
Unbox turns that messy brain dump into structured information:
✅ Tasks
📅 Events & meetings
🔔 Reminders
👤 People
💡 Ideas
❓ Uncertainties
The important part is that Unbox doesn't pretend to know things that weren't said.
If something is ambiguous, it can surface that uncertainty instead of quietly making something up.
The user stays in control: they can review, edit, delete, or confirm what Unbox understood.
The goal isn't to create another complicated productivity system.
It's to make capturing thoughts almost effortless.
Speak first. Organize later.
## Demo
First visit → Name → Voice capture → Transcript → AI organization → Dashboard → Actions
## Code
GitHub: https://github.com/ivy1o1/Unbox
## How I Built It
Voice Brain Dump
↓
Whisper
↓
Transcript
↓
Gemma
↓
Structured JSON
↓
Pydantic
↓
User Review
↓
SQLite
↓
Dashboard
Unbox uses React + TypeScript + Vite on the frontend and FastAPI + SQLAlchemy + Pydantic + SQLite on the backend.
Whisper handles speech-to-text, while Gemma 3 1B Instruct runs locally through llama.cpp to extract tasks, events, reminders, people, ideas, and uncertainties.
The key principle is:
AI proposes. The user decides.
Open-weight AI also keeps the system more inspectable and makes local processing possible for sensitive personal data.
## Why Does Open Innovation Matter?
Brain dumps can contain highly personal information. Using open-weight models makes local AI processing possible without relying entirely on closed APIs.
It also lets me understand and control the full AI pipeline—from model and prompt to validation and uncertainty handling.
Prize Categories
Best Use of Gemma — Unbox uses Gemma 3 1B Instruct locally through llama.cpp as its semantic extraction engine, turning brain-dump transcripts into structured tasks, events, reminders, people, ideas, and uncertainties.
Top comments (0)