GemmaJudges

Judges are usually encumbered by the number of submissions they have to read. GemmaJudges runs three local Gemma 4 AI models that evaluate code, presentation, and creativity.

Comment

Idea

To create a group of custom prompted AI judges that judge individual values in a maker/hackathon competition.

Methods

We use ollama to create 3 differently system, prompted Gemma 4 AI's.

Here are the three different AIs

  1. codeJudgeAI - This ai is prompted to judge the code on conciseness, clarity, efficiency, and ability to function.
  2. creativityJudgeAI - This ai is prompted to judge the idea of the application/creation on creativity, realism, and other factors.
  3. presentationJudgeAI - This AI utilizes the Gemini live transcription API to transcribe a presentation's quality on conciseness, clarity, and engaging speech.

Conclusion

The biggest issue we had was

  1. The Gemini Transcription Servers

  2. Flask connection to a deprecated app on android that would've let us upload everything through to app and then the app sends it to our python server on a more powerful laptop/desktop. However the google transcription did not work well with the uploading of voice files. So we fixed that with a FastAPI + React solution.

Built With

Share this project:

Updates

Submission history