Skip to main content

About the project

Shikayat turns messy student complaints into something a campus can act on. A student writes in English, Hinglish or Hindi and can attach a photo. Gemma 4 (gemma-4-26b-a4b-it, via the Gemini API) picks the right bucket, writes a neutral public summary, judges severity, and checks whether the issue is already open — merging it if so, so the report count rises instead of a duplicate being created. When something an authority actually needs is missing, like which hostel and room, Gemma asks the student back in their own language rather than filing a useless ticket. A public board ranks issues by priority (reports × severity), and each authority sees only its own queue.

We also rebuilt the workshop’s agent harness: per-agent tool sets, image input, terminal tools, and pause-for-a-human so an agent can ask a question mid-run and resume the same conversation. That took a complaint from 4 Gemma calls down to 2. 36 tests, all on a fake model, so they run with no API key. MIT licensed.

Built with

gemma

AI tools

Coded withClaude CodeOpenCode

Prize Categories Entered

  • Deepmind Google

    Best Use of Gemma 4

    Submitted
  • DigitalOcean

    Best Open-Source AI Project

    Submitted