HungryHive

Find the meal that fits the whole hive.

Comment

🍴 HungryHive – Smart Group Dining Assistant

GitHub Repository

💡 Inspiration

The inspiration for HungryHive came from a real, recurring, and frustrating problem: the endless "where should we eat?" debate.
We've all been in that group chat, stuck in a 20-minute loop of conflicting cravings ("I want spicy Indian", "No, I want light Italian"), different hunger levels, and varying locations.
This group indecision often ends in frustration or a mediocre compromise.

We wanted to build a smart, objective tool that could take all those messy human preferences and turn them into a single, optimized, and data-driven decision.

🍽️ What it does

HungryHive is a full-stack web application designed to find the perfect restaurant for a group.

Key Features

  1. Create or Join a Group

    • Users can create a new group, generating a unique ID, or join an existing group using its ID.

  2. Submit Preferences
    Each member fills a simple, interactive form including:

    • 🌍 Location: Auto-fetched via the browser’s Geolocation API.
    • 🍛 Cuisine cravings: (1–5 sliders for Indian, Italian, Mexican, etc.).
    • 🍽️ Hunger & spice level: Radio buttons for clarity.
    • 🥦 Dietary needs: Veg / Non-Veg.
    • 🍹 Drink preferences: Mocktails, juices, cocktails, or alcohol.

  3. Real-Time Group Sync

    • Using Firebase Realtime Database, all members can see who has joined and submitted their preferences — live.

  4. Backend Analysis (Flask + AI)

    • When “Find Restaurants” is clicked, all preferences and locations are sent to the backend.
    • The Python Flask backend calculates the group’s geographic midpoint and uses K-Means Clustering (scikit-learn) to detect subgroups with similar tastes (e.g., “the spicy/hungry cluster”).
    • This structured data is fed to Google Gemini API, which recommends the top 1–2 cuisines that best satisfy the entire group.

  5. Restaurant Search

    • The backend queries SerpAPI (Google Maps engine) using the recommended cuisines and group centroid to find real nearby restaurants.

  6. Vote on Results

    • The backend returns the top 5 restaurant options, which are displayed for live voting via Firebase.

  7. Tie-Breaker

    • A fun “Pick Randomly” button acts as a tiebreaker, selecting one of the top-voted options.

🏗️ How we built it

Frontend:

  • HTML5, CSS3, and vanilla JavaScript
  • Firebase Realtime Database SDK for live syncing of members and votes

Backend:

  • Python Flask framework

Machine Learning:

  • scikit-learn (KMeans, StandardScaler, PCA) for clustering and pattern analysis

Generative AI:

  • Google Gemini API (google-generativeai) to interpret clusters and produce human-readable recommendations

Restaurant Data:

  • SerpAPI (Google Maps) to find actual restaurants near the group’s centroid

Deployment:

  • Frontend: Static Site hosted on Render
  • Backend: Flask Web Service deployed on Render
  • Version Control: Git & GitHub

⚙️ Challenges we ran into

  1. Finding a Viable Restaurant API

    • The Google Places API required billing, so we pivoted to SerpAPI, which offered accessible and reliable search results.

  2. CORS Errors

    • Connecting a deployed frontend and backend caused cross-origin issues. We fixed this by configuring flask-cors and defining environment variables on Render.

  3. Firebase Path Errors

    • Firebase paths can’t contain . (periods), which caused our vote keys (based on coordinates) to fail.
    • We sanitized them by replacing . with _ before saving.

  4. Browser Caching

    • Changes to app.js weren’t updating due to aggressive caching. We learned to clear caches, bump Service Worker versions, and unregister old workers.

  5. Backend Environment Errors

    • We faced environment inconsistencies between local and Render deployments — especially Gemini library version issues and Matplotlib requiring a non-GUI backend (Agg).

🏆 Accomplishments that we're proud of

  • Built a Complete Full-Stack Solution: A seamless frontend, backend, and real-time database integration.
  • Meaningful AI + ML Integration: Combined clustering (ML) and generative reasoning (AI) for intelligent group recommendations.
  • Real-World Impact: Solves a universal pain point — the indecisive “food chat.”
  • Successful Deployment: Connected separate Render services and overcame CORS + environment issues.

📚 What we learned

  • 🔑 Always use environment variables for sensitive keys — never hardcode them.
  • 🌐 Plan for CORS early when connecting multiple deployed services.
  • ⚙️ Manage Service Worker caching aggressively during development.
  • ⚡ Firebase is a simple yet powerful tool for real-time sync without WebSockets.
  • 🧩 Always validate and sanitize data structures exchanged between frontend and backend.

🚀 What's next for HungryHive

  • 💰 Budget & Ambiance Filters: Add price range and occasion (“Casual,” “Celebration”) options.
  • 👤 User Accounts: Move beyond temporary group IDs to persistent profiles and saved preferences.
  • 🍲 Menu Item Recommendation (RAG): Use Gemini with Retrieval-Augmented Generation to recommend actual dishes and portions to reduce waste.
  • 🔗 App Integrations: Deep-link with Swiggy, Zomato, or DineOut APIs for instant order or reservation.
  • 🎙️ Multimodal Inputs: Add voice, image, and chat interactions with Gemini Nano for on-device personalization.

No more endless group chat chaos — HungryHive turns debate into delicious decisions. 🍕🍜🌮

Share this project:

Updates