Smart‑Dine

Streamlined digital waiting room for campus canteens — reducing crowding with AI-driven wait forecasting and a live dashboard.

---

Introduction

Canteen congestion during peak meal hours creates long queues, inefficient staff utilization, and cramped seating. Smart‑Dine transforms the canteen into a digital waiting room by combining live telemetry, an operator-facing Kitchen Display System (KDS), a student-facing dashboard, and AI wait‑time forecasting — reducing physical crowding and smoothing service throughput.

---

Key Features

• Live Dashboard (Bento‑Grid UI): Responsive frontend showing per‑station queue depth, wait estimates, velocity, and seat availability.
• AI Wait‑Time Forecasting (Random Forest): Random Forest regressors predict per‑shop queue length and service time to estimate wait minutes and hourly trends.
• WhatsApp Bot (Twilio): Phone-first interaction channel for registration, joining queues, live status, and push notifications when orders are ready.
• Kitchen Display System (KDS): Staff interface to view active queue, serve orders, and log fulfillment (integrates with the ML data sink).
• Live Seat Tracking: Best‑effort occupancy tracking in the database to show live seat availability on the dashboard and via WhatsApp.

---

Tech Stack

• Backend & AI: FastAPI, SQLAlchemy / psycopg2, scikit‑learn (RandomForest), pandas, joblib, matplotlib / seaborn for diagnostics
• Frontend: Vanilla JS, HTML/CSS (Bento‑Grid inspired dashboard), Netlify hosting for static UI
• Cloud Infrastructure: PostgreSQL (Neon/Render/Postgres), Render for API hosting, Twilio (WhatsApp) for messaging

---

System Architecture

Smart‑Dine uses a 3‑tier decoupled architecture:

• Client Tier (Frontend & WhatsApp): Student dashboard (Netlify) and the WhatsApp bot provide read/write interactions and receive notifications.
• Application Tier (FastAPI): Business logic, queue SLA engine, KDS endpoints, authentication flows, and the AI prediction bridge (deployed to Render or similar).
• Data Tier (Postgres): Normalized tables for active_queue, history_log, students, student_phones, menu_items, and shop_settings. The same history log is the ML training sink.

Communication is RESTful between frontend and FastAPI; Twilio webhooks post incoming WhatsApp messages to the /api/whatsapp endpoint; ML code reads from history_log for training and forecasting.

---

Installation & Setup

Prerequisites: Python 3.11+, Git, PostgreSQL (or a hosted DB), and accounts for Twilio and hosting (Render / Netlify).

1. Clone the repo and create a virtual environment

git clone <repo-url>
cd Smart-Dine
python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

2. Environment variables (.env)

Create a .env in the backend/ folder (and copy to root if you run scripts there). At minimum set:

• DATABASE_URL — Postgres connection string (postgresql://user:pass@host:port/dbname)
• TWILIO_ACCOUNT_SID — Twilio account SID
• TWILIO_AUTH_TOKEN — Twilio auth token
• TWILIO_WHATSAPP_NUMBER — Twilio WhatsApp-enabled number (e.g. whatsapp:+1234567890)

Optional / recommended:

• RENDER_SERVICE_URL — If using Render environment variables for production
• SECRET_KEY — App secret for future auth features

3. Initialize the database (example uses the helper)

python backend/database.py
# or import and run initialize_cloud_database() from a Python REPL

4. Run the API server locally for development

cd backend
uvicorn main:app --reload --host 0.0.0.0 --port 8000

Frontend: Serve frontend/index.html locally or deploy the frontend/ folder to Netlify; change BASE_URL in frontend/script.js to your API host.

---

ML Implementation

• Model: RandomForestRegressor (scikit‑learn) is used for two targets: predicted service_duration and queue_length given day_of_week, hour_of_day, and minute features.
• Training & Prediction: backend/ml_model.py reads history_log from Postgres, trains separate RF models, and produces minute/quarter‑hour forecasts plus an hourly graph for UI display.
• Validation: The repository includes a generate_residual_plot() utility that saves residual plots. Residual analysis is essential to detect bias, heteroscedasticity, and to validate whether a Random Forest is adequate before moving to more complex models.

---

Deployment Notes

• Host the FastAPI backend on Render (or any containerized host). Configure environment variables in the host's dashboard.
• Deploy the frontend to Netlify and point to the backend API. Ensure CORS entries in backend/main.py allow the Netlify domain.
• Configure Twilio WhatsApp webhook to POST to /api/whatsapp on your deployed API URL.

---

Future Scope

• Digital Payments: Integrate UPI / card payments and include payment slots to speed fulfillment.
• Deep Learning Upgrades: Upgrade forecasting to LSTM / Temporal CNN models using per‑order telemetry and embeddings for better long‑horizon predictions.
• IoT & Hardware Expansion: Add QR‑based seat sensors, BLE beacons, or Raspberry Pi counters for robust seat occupancy and real‑time KDS triggers.
• Operational Analytics: Add staff dashboards, SLA alerts, and automated dynamic staffing recommendations.