Back to all updates

2 months ago

⏱️ The Final Sprint: Polishing Your Gemini Live Agent for the Win

Hey everyone, 

As we approach the deadline (only 3 days to go!), the "feature creep" is real, and the deadline is looming: March 16th at 5:00 PM PT.

At this stage, the best projects shift from building to storytelling and stability. Here are some "pro-tips" to ensure your multimodal agent stands out to the judges in these final days.

Don’t forget: ALL Projects must meet the following three criteria at minimum: 

1. Leverage a Gemini model

2. Agents must be built using either Google GenAI SDK OR ADK (Agent Development Kit)

3. Use at least one Google Cloud service (deployed on Google Cloud)

🛠️ Technical Polish: Multimodal or Bust

We aren't looking for a basic chatbot. We are looking for Next-Gen Agents. In these last few days, audit your project against your chosen category:

• Live Agents: Can I actually interrupt the agent? If you haven't tested the "natural conversation" flow with the Gemini Live API or ADK lately, do it now.

• Creative Storytellers: Is the output truly interleaved? Ensure your agent isn't just dumping text then an image, but weaving them into a single, cohesive media stream.

• UI Navigators: Does it actually "see"? Make sure your agent is interpreting screenshots or recordings to drive its actions on the screen.

☁️ For the "Non-Devs": Proving Your Google Cloud Deployment

We know the "Proof of Deployment" can feel intimidating if you used AI to help you code and aren't used to navigating the Google Cloud Console. Don't panic. Here is the easiest way to prove your backend is on GCP without being a sysadmin:

1. The "Cloud Console" Walkthrough: Open your screen recorder. Show us your browser tab where you are logged into the Google Cloud Console. Click on the service you are using (like Cloud Run, App Engine, or Vertex AI). If you can show a "Logs" screen with text moving—or a dashboard showing "Green" checkmarks for your deployment—you’ve nailed it.

2. The Code Shortcut: If a recording is too much, go to your Code Repo. Find the specific file where you call the Google API (it usually mentions vertexai or google-cloud). Grab the link to those specific lines of code and put it in your submission form.

3. Host the Proof: You can upload your .mp4 proof video directly into your code repo. It makes it incredibly easy for you to paste into the submission form for judges to find. 

🎥 The Demo Video: Keep it Real

You have less than 4 minutes. Do not waste time on a long intro.

• No Mockups: The judges want to see the agent working in real-time.

• The Pitch: Clearly state: What problem did I solve? Why does this matter?

• It’s okay to use an AI voiceover – any narration is better than none. 

• The "Wow" Moment: Show the multimodal interaction immediately. If it's a Live Agent, talk to it. If it's a Storyteller, show the mixed-media flow.

• Rehearse your script before you record to ensure your 4 minutes are meaningful. 

💎 The "Judge's Favorite" Checklist

• Architecture Diagram: Make it easy for the judges to find - add it to your image carousel or file upload on Devpost. We want to see how Gemini connects to your GCP backend at a glance.

• README: Make sure your "spin-up" instructions are clear. If a judge can't figure out how to run your code, they can't appreciate your genius.

• Bonus Points: If you have a couple of hours to spare, write a quick blog post about your journey using #GeminiLiveAgentChallenge. It’s the easiest way to grab those extra points. 

🏁 Final Reminder

The deadline is March 16th @ 5pm PT.

Good luck!