Gemini 3 Hackathon Singapore
Gemini 3 has arrived. Google's most intelligent model yet. With state-of-the-art reasoning to help you learn, build, and plan anything.
We’re looking for the most ambitious applications you can build with the Gemini API across Google AI Studio and Google's new agentic development platform, Antigravity.
Compete to win 100k in Gemini API credits and an exclusive 30-minute virtual call with the founders of Google’s AI Future Fund.
Note: This event is fully in-person in Singapore. Space is limited. You will need to present a ticket QR code and photo ID to gain access. Please do not make the trip if your application hasn't been approved. Maximum team size is four. All attendees must submit an application for approval.
Agenda
08:30 AM - 9:30 AM: Registration & Breakfast
Check-in, swag collection, and team formation/mingling.
9:30 AM - 10:15 AM: Welcome & Workshop
Opening remarks from Google DeepMind and 65labs. Brief Google AI Studio & Antigravity workshop for warmup.
10:30 AM: Hacking Begins
Ready, set, hack!
12:30 PM: Lunch
Working lunch to maximize hacking time.
05:30 PM: Code Freeze & Submissions
Hard stop for coding. Teams submit GitHub and video links.
05:30 PM: Dinner & Pre-Judging
Dinner while judges select the top 6 projects.
06:45 PM: Finalist Presentations
Top 6 teams present on the main stage to all attendees & judges.
08:00 PM: Winner Announcement
Awards for 100k API credits & Founders call. Closing remarks.
08:30 PM - 09:00 PM: Exit
🚀 Hackathon Tracks
1. Infinite Context & Multimodal Insight
Focus: SOTA Reasoning & Multimodality
Description:
Data doesn’t live in a vacuum—it exists in video, audio, massive text logs, and code repositories. This track challenges you to utilize Gemini’s massive 1M+ context window to ingest, analyze, and reason across disparate modalities simultaneously. We are looking for solutions that extract high-fidelity facts from complex, "messy" real-world data where traditional models fail.
Examples (what not to build!):
The Legal Eagle: An app that ingests hours of court video footage, deposition transcripts, and scanned evidence files to identify contradictions and generate a timeline of events.
The Diagnostic Detective: A tool for doctors that analyzes patient history, X-ray/MRI imagery, and messy doctor’s notes to suggest differential diagnoses with cited reasoning.
The Black Box decoder: A DevOps agent that reads hours of server logs and screen recordings of dashboard alerts to pinpoint the exact root cause of a crash.
2. The 10x Engineer Agent
Focus: Agentic Coding
Description:
Move beyond simple "tab-to-complete." This track is for building agents that understand the entire software development lifecycle. Use Gemini to refactor legacy codebases, architect new systems, or autonomously debug complex errors. Crucially, your submission must leverage safety evaluations to ensure the code generated is secure and production-ready.
Examples (what not to build!):
Legacy Lifter: An agent that takes a repo written in COBOL or Python 2.7 and refactors it into modern Go or Rust, including writing new unit tests.
Security Sentinel: A bot that autonomously scans a GitHub repository for vulnerabilities, generates a fix, and opens a Pull Request with a detailed explanation.
Text-to-SaaS: A system that takes a natural language description of a web app and deploys a fully functional prototype to the cloud, handling backend logic and frontend design.
3. Enterprise Orchestration & Autonomous Actions
Focus: Advanced Tool Use & Planning
Description:
AI needs hands. This track focuses on "Function Calling" and planning. Build an agent that can break down ambiguous high-level strategies ("Improve customer retention") into concrete, multi-step workflows involving external APIs (CRMs, Email, Databases) with minimal prompt tuning. The goal is precise, reliable execution.
Examples (what not to build!):
The Supply Chain Commander: An agent that notices a weather delay (Weather API), re-routes a shipment (Logistics API), and emails the customer an update (Gmail API) automatically.
The Personal Chief of Staff: An assistant that plans a full business trip—booking flights, hotels, and dinner reservations based on your calendar and dietary restrictions—asking for confirmation only before payment.
Customer Ops Auto-Pilot: A support bot that doesn't just answer questions but performs actions, like processing refunds or updating shipping addresses, by authenticating into the company database.
4. The Generative Media Studio
Focus: GenMedia (Gemini TTS, Lyria, Nano Banana, Veo)
Description:
Unleash the creative potential of AI. This track is for applications that generate compelling audio, video, and music experiences. Whether you are using Veo for video generation, Lyria for music, or Gemini’s hyper-realistic Text-to-Speech, your goal is to build an immersive media application that entertains or educates.
Examples (what not to build!):
Dynamic Storyteller: An interactive education app where history comes alive via Veo-generated video clips and narrated by specific historical figures using Gemini TTS.
Adaptive Game Soundtracks: A video game plugin that uses Lyria to generate background music that reacts in real-time to the player's intensity and gameplay style.
The "Nano Banana" Remix: A viral content generator that takes boring news articles and converts them into catchy, short-form musical video summaries.
5. Deep Research & Knowledge Synthesis
Focus: Gemini Deep Research & Interactions API
Description:
Information overload is a massive problem. This track utilizes the Deep Research capabilities to perform recursive, exhaustive investigations. Build tools that can browse, read, synthesize, and report on vast amounts of knowledge, while using the Interactions API to present this data in structured, user-friendly formats.
Examples (what not to build!):
The Thesis Defender: A research assistant that crawls thousands of academic papers to generate a literature review, highlighting gaps in current research.
Market Mover: A financial analyst agent that reads annual reports, news streams, and social sentiment to produce a deep-dive investment memo with interactive charts.
Fact-Check Prime: A journalism tool that takes a breaking news topic, performs deep web verification, and presents a "truth score" dashboard.
6. Real-Time Multimodal Experiences
Focus: Voice & Vision Agents with Gemini Live API
Description:
Speed matters. This track is dedicated to the Gemini Live API, focusing on low-latency, conversational voice and vision interactions. Build agents that can see the world through a camera and converse naturally with the user in real-time.
Examples (what not to build!):
The Co-Pilot Mechanic: A mobile app where a user points their camera at a car engine, and the AI guides them through a repair step-by-step via voice conversation.
Accessibility Now: A wearable-integrated assistant for the visually impaired that describes the environment, reads street signs, and identifies obstacles instantly.
Interview Coach Live: A desktop app that watches you practice a presentation and gives real-time audio feedback on your body language and speaking pace.
🏆 Prizes & Allocations
Total Prize Pool: 100,000 Gemini API Credits
Power your startup, scale your experiments, and run the heavy models on us.
🥇 1st Place: Grand Prize
Prize: 30,000 Credits
Awarded to the best overall application that demonstrates technical excellence, real-world utility, and masterful use of Gemini capabilities.
🥈 2nd Place: Runner Up
Prize: 20,000 Credits
Awarded to the project with exceptional execution and high potential for impact.
🥉 3rd Place: Second Runner Up
Prize: 10,000 Credits
Awarded to the project with strong technical implementation and a polished user experience.
🎖️ Honorable Mentions (x5)
Prize: 6,000 Credits each (30k Total)
Five awards dedicated to specialist excellence. These will go to the best projects in specific verticals (e.g., "Best Use of Audio," "Best Agentic Workflow," "Best UI/UX").
🍌 The "Most Bananas" Hack
Prize: 10,000 Credits
Awarded to the most creative, unconventional, or wildly entertaining use of the technology. We are looking for the "weird but wonderful" idea that pushes the boundaries of what we thought the models could do.
Judging & Judges
We'll use a Gavel judging system to select the top six projects for the finals. Finalists will present live on stage. Thank you to the judges for offering their time and expertise.
(from left to right)
Manikantan Krishnamurthy - Developer Relations Head @ Google - APAC
Sherry Jiang - Co-Founder peek.money | codewithai.xyz
Evan You - Founder voidzero.dev
Thor Schaeff - DX @ Google DeepMind
Ivan Leo - Agents at Manus
Agrim Singh - Co-Founder niyam.xyz
Natalie Khoo - Vibing at Manus