BawarchiAI 🍲 — Zero-Waste Canteen Planner

BawarchiAI is an end-to-end, production-grade AI platform designed to minimize food waste in canteens and institutional kitchens. By blending machine learning predictions with historical baseline telemetry, the platform optimizes preparation quantities, registers leftover surplus, and routes surplus food to local NGOs.

🔗 Live URLs

• Live Web App: https://bawarchi-ai-rust.vercel.app
• Go Backend Server API: https://bawarchiai-backend.onrender.com
• Python FastAPI ML Service: https://bawarchiai-ml.onrender.com

---

📈 System Metrics & Performance

• In-Sample Model Fit ($R^2$): 99.92% (MAE = 1.17 kg | RMSE = 1.53 kg)
• Out-of-Sample Generalization (5-Fold Cross-Validation $R^2$): 97.88% (MAE = 5.70 kg | RMSE = 7.65 kg)
• Inference Latency: <350ms (hybrid forecast evaluation)
• Ecological & Operational Impact:
  • Food Saved: 5,698 kg (includes 15% AI-prevented waste + NGO redistribution)
  • Meals Redistributed: 22,793 meals (standardized 250g portion size)
  • $CO_2$ Emissions Prevented: 13,106 kg (EPA/WRI conversion factor of 2.3x)

---

🏗️ Architecture & Data Flow

graph TD
    User[Web Client / Operator] -->|HTTPS| FE[Next.js Frontend: Vercel]
    FE -->|JSON Requests| BE[Go Gin REST API: Render]
    BE -->|SQL queries via PgBouncer Pooler| DB[(Supabase Postgres)]
    BE -->|Predict / Train requests| ML[FastAPI ML Service: Render]
    ML -->|Trains XGBoost Regressor| Models[(Models Directory)]
    Models -.->|Reads metrics & models| ML

Loading

1. Hybrid Forecasting Engine

The ML service uses a blended forecasting algorithm to combine statistical averages with predictive regressor weights: $$\text{Expected Consumption} = 0.7 \times \text{XGBoost Prediction} + 0.3 \times \text{Historical Canteen Average}$$

Recommendations are scaled using Confidence Multipliers based on historical sample support ($N$) and Coefficient of Variation ($CV = \sigma / \mu$):

• High Confidence ($N \ge 20$, $CV < 0.15$): 1.05x safety buffer.
• Medium Confidence ($N \ge 8$, $CV < 0.30$): 1.10x safety buffer.
• Low Confidence (Fallback): 1.18x safety buffer (prevents student shortages).

---

🛠️ Technology Stack

Component	Technology	Description
Frontend	TypeScript, Next.js 15, TailwindCSS, Recharts	Dynamic dashboards, Recharts feature importances, and NGO logs
Backend	Go (Golang), Gin Gonic, pgx/v5	REST API, database connection pooler, middleware CORS routing
ML Service	Python 3.10+, FastAPI, XGBoost, Pandas, Scikit-Learn	Training pipeline, 5-Fold Cross-Validation, Feature Importances
Database	PostgreSQL (Supabase), PgBouncer	Serverless SQL hosting, transaction pooling using Simple Protocol
Infrastructure	Docker, Docker Compose, GitHub Actions	Multi-container local orchestration and CI/CD triggers

---

📂 Project Structure

├── backend/               # Go REST API Server
│   ├── cmd/server/        # Entry point (main.go)
│   ├── internal/          # Core modules (forecast, surplus, impact, donation)
│   └── Dockerfile
├── frontend/              # Next.js App Router Frontend
│   ├── app/               # React Page Views & Layouts
│   ├── components/        # Redesigned Green Zero-Waste Components
│   └── lib/api.ts         # Non-cached API client fetching
├── ml-service/            # Python FastAPI Machine Learning service
│   ├── app/               # Models training, encoders, features preprocessing
│   └── Dockerfile
└── docker-compose.yml     # Local orchestration stack

---

⚡ Local Setup

1. Prerequisite

• Install Docker and Docker Compose.

2. Launch Local Environment

Clone the repository and spin up the backend, database, and machine learning services:

git clone https://github.com/AkshajSonar/BAWARCHI.AI.git
cd BAWARCHI.AI
docker compose up --build -d

3. Seed Database & Train Model

In a separate terminal, seed the database with canteens training data and trigger model retraining:

# Seed local Postgres
docker exec -i bawarchiai-postgres-1 psql -U plateai -d plateai < path/to/seed.sql

# Trigger retraining on local Go server
curl -X POST http://localhost:8080/forecast/retrain

4. Run Frontend Development Server

Navigate to the frontend directory, install dependencies, and run Next.js:

cd frontend
npm install
npm run dev

Open http://localhost:3000 to view the application.