Roolin：首演前預演突發狀況

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle.

What I Built

It is 19:03 on the night of the school showcase. The soloist is not in the wings. The host is looking at you, the audience is looking at the curtain, and you have about a minute to say something to the crew that does not make it worse.

Roolin is a local 3D rehearsal room for that minute.

It gives teachers, stage managers and community-theatre organisers somewhere to practise responding when a microphone fails or a performer is late. A theatre volunteer should not need to understand AI or programming to practise a difficult moment. Once the local setup is complete, the app guides them through the running order, their response and the next thing to check.

Bring your running order, review the cues proposed by Gemma, and explore a representative theatre—from the audience to backstage. At a cue, choose a disruption and write what you would actually tell the team. Gemma responds to your words with coaching and a concrete next check.

For example, when the soloist is late, you can practise briefing the crew and holding the show. Accepting that hold moves the current and later cues by five minutes while preserving the original schedule. Using a reported tested spare microphone leaves the times unchanged.

Preparation helps you work backwards from the first cue to decide when setup should begin.

The potential paid direction is team plans for drama departments and small production companies: shared running orders, collaborative rehearsals and crew-ready handoffs. Those are future possibilities. I’ve spoken with different people about whether this is something they would pay for; those conversations are exploratory, rather than confirmed purchases.

Demo

Watch the Roolin demo on YouTube

“The Show Before the Show” follows a fictional theatre volunteer facing a late soloist. Product scenes show the real Roolin app running with local Gemma and demonstrate how the rehearsal flow works.

Code

Roolin on GitHub

The application is MIT-licensed and runs locally. The README covers installation, model setup and limitations.

How I Built It

I built the theatre with Three.js, alongside a JavaScript interface and a local Node server.

Gemma 4 E4B, running through Ollama, handles two jobs: turning plain-text show notes into proposed cues with source quotations, and coaching written responses to disruptions.

Application code checks cue times, quoted text and valid schedule changes. The organiser reviews the results and approves changes.

All 18 automated tests pass, and separate browser walkthroughs exercised real local inference. AI tools assisted development and demo production.

This is a single-user prototype with two simulated disruptions. It does not detect equipment failures or certify readiness. Session data disappears on refresh; export it first.

Why Does Open Innovation Matter?

A lot of useful technology comes with another payment or subscription. Sometimes, you can build what you need yourself using open tools and a computer you already have at home.

That is what I wanted to explore with Roolin. With suitable hardware, open weights and local inference make it possible to create a rehearsal tool without paying a cloud provider for every practice session. Hardware, electricity and initial setup still have a cost.

The person using it does not have to be the person who built it. A theatre volunteer can focus on what they would say to the crew and how to respond to a disruption, while the app provides a place to practise and review the consequences. Making those tools available to people who are not tech savvy is part of the value I want to test.

Local inference also lets rehearsal notes stay on the organiser’s computer. The application code, prompts and validation rules are available for others to inspect and adapt to their own productions.

Prize Categories

Best Use of Gemma: local Gemma powers source-linked cue drafting and response coaching—the two language tasks at the heart of Roolin.