Back to all updates

2 months ago

Submissions are Locked! šŸ”’

First and foremostā€”take a breath. The hard part is over. Thank you for the incredible amount of time, late-night coding, and creative energy youā€™ve poured into the Gemini Live Agent Challenge. Weā€™ve been watching the projects roll in, and the level of multimodal innovation is truly inspiring.

āøļø Submissions are locked in

  • No More Edits: Please refrain from making any changes to your linked code repositories.

  • Keep Building (Separately): If youā€™re excited to keep evolving your agent beyond the hackathonā€”which we love to see!ā€”please fork your repository. Continue your new work in the fork, but keep the original repo exactly as it was at the deadline so judges can review the correct version. 

  • The Project Gallery: We will be reviewing all projects and plan to open the project gallery in just a few days so you can see what your peers have built!

āš–ļø The Judging Process: What Happens Now?

Our panel is diving into your work using the two-stage evaluation process outlined in the official rules.

A Note on Testing: Because of the high volume of incredible entries, judges will prioritize your demo videos. While they may test live backends, they are not required to do so for every projectā€”this is why your 4-minute demo and your "Proof of Deployment" are so critical!

Stage One: Baseline Viability (Pass/Fail)

We first ensure that the submission meets the technical "must-haves": Does it include a video, code repo, and architecture diagram? Does it reasonably address one of the three challenges using Google Cloud and Gemini?

Stage Two: The Weighted Evaluation

Submissions that pass Stage One are scored (1-5) based on the following:

Innovation & Multimodal User Experience (40%):

  • The "Beyond Text" Factor: Does the project break the "text box" paradigm? Is the interaction natural, immersive, and superior to a standard chat interface? Does the agent "See, Hear, and Speak" in a way that feels seamless?

  • Category Execution:

    • (For Live Agent Submissions): Does the agent handle interruptions (barge-in) naturally? Does it have a distinct persona/voice?

    • (For Storyteller Submissions): Is the media interleaved (text, image, and audio) seamlessly into a coherent narrative?

    • (For UI Navigator Submissions): Does the agent demonstrate visual precision (understanding screen context) rather than blind clicking?

  • Fluidity: Is the experience "Live" and context-aware, or does it feel disjointed and turn-based?

Technical Implementation & Agent Architecture (30%):

  • Google Cloud Native: Does the code effectively utilize the Google GenAI SDK or ADK? Is the backend robustly hosted on Google Cloud (Cloud Run, Vertex AI, Firestore)?

  • System Design: Is the agent logic sound? Does it handle errors, API timeouts, or edge cases gracefully?

  • Robustness: Does the agent avoid hallucinations? Is there evidence of grounding?

Demo & Presentation (30%):

  • The Story: Does the video clearly define the problem and the solution?

  • The Proof: Is the architecture diagram clear? Is there visual proof of Cloud deployment in the video or submission materials?

  • The "Live" Factor: Does the video show the actual software working (not just mockups)?

šŸ’¬ Feedback & Community

With so many groundbreaking projects, we are unable to provide individual feedback on every submission. However, some of the best insights come from those who were in the trenches with you!

We highly encourage you to head over to the Devpost Discord, share your project links once the gallery is live, and swap feedback with your fellow developers.

Good luck!