Growth Garden
Tell it what you want to get better at. It gives you tasks. Finish them and your planet grows. 1st of 135+ projects at Australia’s largest student hackathon.
Play the original loop
This is the actual UniHack build, moved into the page. Bruce calls OpenRouter for a live set of tasks, then the original reward loop takes over.
What changed
The hackathon version sent Bruce's prompt through Langflow and used DataStax for persistent users, tasks and gardens. This page keeps the actual interaction in place, makes a fresh server-side OpenRouter call, and holds the demo state only for this visit. Failed calls show an error; there is no canned response.
The sound
The hackathon build did not use one long ambience recording. It looped this 55-second track, then scheduled owl calls every 5 to 15 seconds, one of five bird and rain clips every 8 to 18 seconds, and bubbles every 20 to 30 seconds. Revealing, picking up and placing a reward each fired a separate sound.
This version keeps the loop and the three reward sounds. The player and the Sound control above use the same audio, so you can start it here and carry on with the demo. It sits at 12% volume under Bruce.
Happy Relaxing Loop
SergeQuadrado · 55 seconds · 192 kbps MP3
What it is
Habit apps usually reward you with a tick in a box. Growth Garden gives you a planet instead. You tell Bruce what you want to improve, he turns it into concrete tasks, and every task you finish unlocks a new 3D object for your world. The garden is the streak.
The reward first appears glowing beside the planet. You pick it up, drag it onto the world, and it settles against the curved surface. That small bit of ownership made finishing a task feel different from clearing a checkbox.
My part
- Bruce. I built the assistant in Langflow. Goals go in, tasks come out, and the conversation keeps context across the session by chaining prompts rather than fine-tuning anything. No fine-tuning was the right call for a weekend: the model never needed to know about gardens, it just needed to be good at breaking a goal into steps.
- The data. Users, habits, tasks and the state of every garden live in DataStax. I built the data model and the queries, most of them inside the same Langflow flows, so the garden updates the moment a task is ticked off.
- The pitch. I led the presentation. The order that wins a hackathon is pitch, demo, code, and I planned it in that order: the story first, then a demo that couldn’t fail live, then the architecture for the judges who asked.
- The physics. Reviewed the placement system with the team and helped keep it at a smooth frame rate while the planet filled up.
Result
- 1st of 135+. Judged across entries from Australia and New Zealand. I was the youngest UniHack winner on record.
- What I learned. Judges never see your code. They see whether the demo works and whether they understood it in the first thirty seconds. We spent Sunday morning cutting features so the demo would survive being run by a nervous person on a projector.
The team
Street Corn was mostly the DeepNeuron neural cellular automata team, pointed at a different problem for a weekend.
- Joshua Riantoputra
- Angus Bosman
- Chloe Koe
- Keren Collins
- Luca Lowndes
- Nathan Culshaw
Stack
Next.js, React, TypeScript, Three.js, DataStax, Langflow, OpenRouter