Loop In

Tagline / one-line summary Loop In — jam like a musician without being one. A shared beatbox circle where an AI bandleader pulls anyone into the song.

By team Attempt 5 (Hera Choi)

Inspiration

I started out wanting to build a tool for myself — a songwriter trying to capture ideas before they vanish. Partway through I realised the more interesting version wasn't about me. At a beatbox workshop I'd been to, almost nobody could really beatbox, but it didn't matter: someone laid down a clumsy beat, the lead said "anyone got hi-hats?", and a nervous person added a little tss-tss-tss. Suddenly there was a groove nobody could've made alone. The joy of jamming together has almost nothing to do with skill — but most people never reach it, because the skill barrier stops them at the door. Loop In removes that barrier for non-musicians, without pretending to hand them years of practice.

What it does

Loop In is a shared audio loop you join like a beatbox circle. Open it and a groove is already playing, so adding to it feels like joining, not starting from scratch. You record a short sound — voice, beatbox, hum, tap — and it loops back in as your layer, tagged with your name. An AI bandleader reads what the groove has and what it's missing, then sizes the next invitation so anyone can say yes: a first-timer gets "just say your name," a fuller loop gets "that bassline's lonely — who's got something bright on top?" When it fills up, the AI producer reacts to the finished jam and drops it as a release card — a styled single title and liner notes naming the contributors. It's not meant to sound like a hit. It's meant to sound like you, made together.

How we built it

Mobile web app (React ) so any judge can join from a phone — no install. Audio runs on Tone.js / Web Audio: a looping seeded bed, live mic capture, and contributions scheduled to the transport so they lock to the groove. The AI bandleader runs on Manus — it takes the live loop state (layer count, what's present, what's missing), the genre, and the last contributor's name, and generates the next invitation, with a hardcoded prompt list as fallback so the demo never stalls. Built solo - time-blocked hat-switching, validating ideas with LLM doing research.

Challenges we ran into

The biggest time sink was tooling, not design. My first coding agent kept breaking the build because the existing loop state wasn't baked in, and I lost real time babysitting repeated errors. I also took a UI detour — generated a design in Figma, exported it for the agent to build on, and that's where it fell apart — before going back to Manus, which coded the working version cleanly. The hardest calls were cuts: I removed an "instrument" feature that recorded the mic then threw the audio away to play a preset, because it contradicted the whole point (your real sound becoming the song). And I deliberately didn't build perfect quantization — some looseness is what makes it feel human.

Accomplishments that we're proud of

Most of all, the pivot: I let go of "make a tool for me, the songwriter" and turned it into something that gives non-musicians the part of music that's actually joyful — jamming together. Coming from real production tools — DAWs with endless knobs and menus — I'm proud of how much I stripped away: I had to un-learn the interface I know and ask what the experience of jamming actually is, then hide everything else. The result is a complete, joinable jam built solo in a day, where someone with zero musical background can contribute in seconds and feel like they made something. And the AI genuinely drives it — it reads the room and meets each person where they are, instead of being decoration.

What we learned

Under time pressure, subtraction beats addition — a feature that fights your core idea is worse than no feature. Technically, I learned the browser-audio stack hands-on: Tone.js transport scheduling, mic capture, HTTPS constraints on mobile, and where AI adds real intelligence (reading state, meeting a person where they are) versus where it's just dressed-up if-statements. I also learned where different AI coding tools help and where they cost you.

What's next for Loop In

• Hum-to-instrument: pitch detection (Basic Pitch) so a hummed or beatboxed line becomes a real instrument — keeping the human performance as the source.
• True multiplayer: shared live sessions and a "pass the mic" handoff, so the connection runs person-to-person.
• Reading confidence from audio: infer how tentative or bold a contribution is and let the bandleader coax the shy, push the bold.
• Richer beds: AI-generated seeded grooves per genre, and a mixed/mastered final single.

Built With

• figma
• manus