GDG Coding Jams
Build with AI. Together.
Share
← Back to the jams
Track 09

Your Own Idea

No menu. No template. The thing you've been daydreaming about.

✅ Drop-in friendly2-hour jamShips in 45 min
💡
The demo

What it looks like when it’s working

Demo video goes here

Demo of Your Own Idea

Drop the 11-char YouTube ID into youtubeId on this track in lib/tracks.ts.

In the room · 45 minutes

What you’ll build

Bring an idea you've been sitting on — write a 1-paragraph PRD, hand it to Antigravity, and ship a working slice in 45 minutes. No starter repo. No menu. Just your spec and your taste.

The moment it clicks

I built the thing that was only in my head.

Things to think about

While you build

  • One paragraph of PRD beats a backlog. If you can't describe it in three sentences, you can't ship it in 45 minutes.
  • Pick the smallest version of the idea that's still recognizable — one core flow, one screen, no settings.
  • Your signature detail matters more here than anywhere else. There's no demo to copy from, so the soul has to come from you.
  • Stuck on prompting? Steal from the other tracks — image gen, RAG, persona design, agent loops are all fair game.
The 2-hour Jam rhythm

How the night flows

Every Coding Jam follows the same five-phase shape. Talking is short, building is long.

  1. 0:00 – 0:15
    Intro

    Grab a name tag and a slice, say hello to the instructor, and meet the people either side of you — you'll be asking them things later, so it pays to know their names now. Low pressure, high creativity, ship something messy. Then a quick word on what the room is building tonight, and you're off.

  2. 0:15 – 0:30
    Credits & setup

    The instructor shares the Google Cloud credits with the room. Everyone gets their environment ready — whichever tool they brought. Nobody should be installing anything once the build starts.

  3. 0:30 – 1:35
    Build

    Heads down, and enjoy it — this is the part everyone came for. The agent does the typing; you decide what it's making. Ask the room when you get stuck; someone two seats away hit the same wall ten minutes ago. Want more structure? The codelab walks the spec-driven route — Setup → Plan → Review → Build → API → Verify — but nobody has to follow it.

  4. 1:35 – 2:00
    Share & submit

    Show the room what you made — two minutes, screen-shared from your seat. Half-finished is welcome; the wobbly ones are usually the most interesting. Say what surprised you along the way, and cheer for everyone else while you're at it. Then submit your build before you leave. This is the one thing not to skip — it's what puts your name and your chapter on the showcase, and it takes about two minutes.

Polished version pulls in

Where to take it after the jam

The 45-minute build is the win in the room. These are the ideas you can pull in over the next week — your homework isn’t homework, it’s the polished version.

  • Whatever the at-home version of your idea looks like.
  • Steal patterns from tracks 1-8 — they're reference implementations now.
  • Polish pass: empty states, error states, the one delightful detail.
From the community

Builds shipped from this track

All builds from this track →
Anatomy Atelier
Track 09Universiti Tun Hussein Onn Malaysia
Anatomy Atelier

Anatomy Atelier is an interactive 3D human anatomy and biology learning platform built for medical students, educators, and SPM Biology candidates. It combines real-time WebGL 3D organ specimens with an intelligent RAG search engine powered by Gemini API to query official syllabus textbooks. Learners can isolate organ tissues, simulate biological metrics, and ask physiological questions with instant, verified textbook citations.

I expected RAG textbook ingestion and 3D WebGL rendering to create heavy latency, but pairing Gemini API with structured retrieval delivered instant, context-aware anatomical answers while maintaining a smooth 60fps canvas.

Gemini APIGoogle AI StudioAntigravityCloud Run
VONAGE-AHSAN
Track 09GDG Brooklyn
VONAGE-AHSAN

An voice agent built on Flue framework with Vonage API for voice.

It was nice to develop it.

VONAGE-ADIL_SABIR_AZEEZ
Track 09GDG Brooklyn
VONAGE-ADIL_SABIR_AZEEZ

This project is a real-time voice agent that lets users dial a standard phone number to interact with a custom AI persona. Built for developers looking to integrate telephony with streaming LLM pipelines, it serves as a working boilerplate for handling bi-directional audio streams and system prompts. It's a quick way to see how standard voice APIs stack up against modern conversational AI without needing heavy infrastructure.

I expected setting up telephony WebSockets and call state management to require heavy boilerplate. Surprisingly, Vonage’s Flue framework abstracted away all that complexity out of the box, letting me spin up a fully functional AI voice agent in just a few lines of code.

APO-HAL-EMS
Track 09Michigan, USA
APO-HAL-EMS

APO-HAL-EMS (Agnotic Policy Optimizer / Hardware Abstraction Layer For Energy Management Systems) is an architectural framework and universal governance layer designed for critical infrastructure control systems and energy management systems. ​It provides deterministic validation of automated, AI-driven, and human operator decisions before actions are committed to physical hardware or control networks.

I thought I'd have to do more work by constantly correcting and re-checking that my prompts were being followed with accuracy throughout the build. Gemini was pretty solid and made the process so much smoother, creating much less work for me due to high precision and follow-through.

SensePath
Track 09GDG Brooklyn
SensePath

SensePath is a voice-first accessibility assistant designed for blind and low-vision users. It combines Gemini Live, multimodal vision, and Google Maps Platform to help people navigate the last few meters of a journey—finding entrances, benches, elevators, and understanding their surroundings through natural conversation. The standout feature is its ability to seamlessly switch between navigation, object finding, and immersive scene descriptions while keeping the interaction hands-free and acces

We expected navigation to be the hardest problem, but the real challenge was making AI know when not to speak. Designing an assistant that gives timely, concise guidance without overwhelming or distracting the user turned out to be just as important as the underlying AI itself.

HarmonicBlend
Track 09GDG Brooklyn
HarmonicBlend

HarmonicBlend is an AI-powered, browser-based DJ and stem mashup workstation that allows users to search any song on YouTube Music and turn it into a professional-grade remix in real time. The platform features dual interactive vinyl turntables, scrubbable waveform timelines, 3-band EQs, lowpass/highpass filter sweeps, crossfader controls, and an offline WAV exporter that lets users render and download their completed mixes directly to disk, as well as using Facebook's Demucs ML to create stems.

Using Gemini within Google Antigravity to build this project was a leap in autonomous, agentic software engineering. Its crazy how Antigravity reasoned across full-stack boundaries, making Python Flask backends, setting up PyTorch ML pipelines, and coding complex Web Audio API DSP frontend engines, something I would need months of training and understanding to build.

Pick another track

All independent — start anywhere.

See all jams →