Skip to main content

About the project

💡 Inspiration

Between lectures, college breaks are only 15 minutes long, but over a hundred students rush to the canteen simultaneously. Standing in chaotic, manual lines takes 20 to 25 minutes, forcing students to either skip food or arrive late to their next lecture. We built CanteenRush to give campuses an automated pre-ordering and kitchen queue agent that allows students to pre-order food while walking between classes and pick it up right when it’s hot.

🚀 What it Does

  • Conversational Food Pre-Orders: Students chat naturally with Gemma 4 to browse dishes, check wait times, and place orders.
  • Campus Roster Authentication: Validates college roll numbers against an institutional student database to eliminate fake orders and enforce a strict 1-active-order limit.
  • Real-Time Stock Verification: Gemma actively queries live kitchen inventory via dedicated function-calling tools before committing orders.
  • Live Canteen Kitchen Display: Canteen staff see a real-time order dashboard with queue status, itemized orders, and wait times.
  • Zero-Confusion Pickup System: Generates a Token ID and a secret 4-digit Claim PIN for each order, plus instant student pickup notification alerts when staff marks food ready.
  • Interactive Inventory Management: Staff can toggle real-time item availability (In Stock / Out of Stock) with one-click live UI syncing.

🤖 How We Used Gemma 4 & Architecture

  • Model: Google DeepMind’s gemma-4-26b-a4b-it accessed through the official Gemini API (google-genai SDK).
  • Mixture-of-Experts (MoE): Leveraged Gemma 4’s architecture (26B total parameters, 4B active per token) to achieve near-instant response latencies during peak rush hours.
  • Agent Skill Open Standard: Implemented an open-standard agent skill (skills/canteen-preorder/SKILL.md) equipped with structured tools (check_item_stock, verify_student, check_kitchen_queue, place_canteen_preorder, calculator).
  • UI: Interactive dual-view Gradio interface featuring both the student ordering terminal and the kitchen live operational board.

🏆 Challenges Entered

  • Best Use of Gemma 4: Utilizes gemma-4-26b-a4b-it through the Gemini API with structured multimodal/tool-calling capabilities to power an end-to-end campus productivity system.
  • Best Open-Source AI Project: Built with an open-weight model harness, adheres to the Agent Skill Open Standard, and released under the permissive open-source MIT License.

🔮 What’s Next

  • WhatsApp / Telegram Campus Bot: Deploying the agent directly inside messaging channels so students can order without opening a browser.
  • Predictive Kitchen Load: Using historical rush patterns to auto-tune preparation time estimates during exam weeks and fest days.

Built with

gemmapython

AI tools

Coded withAntigravityClaude CodeGoogle AI StudioGemma 4 26B A4B

Prize Categories Entered

  • Deepmind Google

    Best Use of Gemma 4

    Submitted