FoodForAll
Predict. Rescue. Feed.
When the city's kitchens close, FoodForAll opens.
A late-night food platform that serves restaurant surplus to night-shift workers and routes every unsold meal to NGOs and shelters, before it's unsafe.
🌐 Live Demo · 📂 Repository
📖 About the Project
Food delivery apps stop when restaurants close, but restaurants still have surplus cooked food, and nurses, BPO teams, data-centre staff and security guards still need to eat. Meanwhile, that surplus usually ends up in a landfill.
FoodForAll closes that gap in one night-long loop:
- Plan: a machine-learning model forecasts each restaurant's leftovers before it closes, using live weather, festivals, public holidays and footfall.
- Collect: restaurants send surplus to temperature-controlled dark stores, which check it in at a verified safe temperature.
- Serve: night-shift workers order fresh restaurant food at a fair price, 11 PM – 5 AM.
- Share: anything unsold 2 hours before its safe time ends, or at the 3:30 AM cut-off, becomes a free donation that one verified NGO can claim, collect and deliver to a shelter.
Nothing reaches a landfill.
✨ Key Features
| For | What they get |
|---|
| 🍛 Kitchen managers | A 7-night outlook of expected leftovers with live weather and festivals filled in automatically, plain-English explanations ("Diwali festival over-prep: +46% leftovers"), a closing log and one-click dispatch to a dark store |
| 🏬 Dark-store managers | Live hot-cabinet and fridge temperatures, delivery check-in that turns away unsafe food, a safety countdown for every dish, "Close for the night", and an activity log |
| 🌙 Night-shift workers | A late-night menu with diet filters, popularity and "Only 2 left!" badges, instant reservation at checkout, and live order tracking |
| 🤝 NGO coordinators | An early warning of tonight's expected food before kitchens close, a nearby-hubs map, and claim → pick up → deliver tracking with people-fed logging |
| 📊 Operations lead | Access to every portal, SDG impact analytics, forecast accuracy, and safety drills |
Under the hood:
- 🔒 No double-selling, no double-claiming. Stock is reserved atomically: 25 simultaneous buyers for 5 portions means exactly 5 orders succeed. Only one NGO can claim each donation.
- 🌡️ Food safety enforced automatically. Hot food must stay at or above 63 °C (max 6 h) and chilled food at or below 5 °C (max 24 h). Temperature problems shorten a dish's safe time, and serious ones discard it.
- 🌦️ Zero manual input. Weather comes from Open-Meteo (16-day forecast, refreshed hourly). Holidays and festivals come from a generated 2025–2031 Indian and Karnataka calendar.
- 🗣️ Plain language everywhere. No codes or jargon on any page, and each person only sees the pages they use.
🧭 Try It Yourself
Open the live site and use the Quick sign-in list on the login page. Every account uses the password foodforall2026.
⏳ Note: the hosted version runs on serverless infrastructure. The first visit after a quiet spell can take up to a minute while it wakes up, and sample data resets periodically.
🛠️ Tech Stack
| Layer | Technology |
|---|
| Backend | Python, FastAPI, Pydantic v2, SQLAlchemy 2 (SQLite) |
| Machine learning | scikit-learn (Random Forest), pandas, NumPy |
| Frontend | Jinja2, Tailwind CSS, Alpine.js, Chart.js, Lucide icons |
| Security | Argon2 password hashing, JWT authentication, role-based access |
| Data | Open-Meteo weather API, holidays package (Indian calendar) |
| Hosting | Vercel |
🧠 How the Forecast Works
- Inputs: weather, evening temperature, day of week, festivals, public holidays, restaurant size and dish category.
- Model: a Random Forest trained on a year of nightly history across 7 restaurants, using real Bengaluru weather.
- Explanations: each prediction comes with a likely range and the reasons behind it ("Rain: +32% leftovers").
- Accuracy: tested on 8 weeks of nights the model never saw. Average error is about 4.4 servings per dish category on ordinary nights and about 4.9 on festival nights, against about 6.5 for a simple average.
- NGO promise: NGOs are promised a safe portion of the forecast, with a reserve held back in case the night is busier.
🚀 Getting Started
# 1. Clone
git clone https://github.com/Ron102008/FoodForAll.git
cd FoodForAll
# 2. Install (Python 3.10+)
python3 -m venv .venv
.venv/bin/pip install -r requirements-dev.txt
# 3. Run
.venv/bin/uvicorn app.main:app --reload
Open http://localhost:8000. On first start the app creates its database, seeds a year of sample history, fetches live weather and trains the forecast model (about 20 seconds).
# Run the test suite (34 tests, including race-condition tests)
.venv/bin/python -m pytest -q
# Wipe and reseed the sample data
.venv/bin/python seed_data.py --reset
📁 Project Structure
FoodForAll/
├── app/
│ ├── main.py # FastAPI app, startup, background routing
│ ├── ml/ # forecasting model, calendar, synthetic history
│ ├── services/ # weather, inventory, NGO routing, food safety, impact
│ └── routers/ # API + page routes for each role
├── templates/ # pages: landing, store, restaurant, dark store, NGO, analytics
├── static/ # JS, CSS, icons, vendored libraries
├── scripts/ # calendar + icon generators
├── seed_data.py # one year of sample history + tonight's scenario
└── tests/ # pytest suite
🌍 Impact
FoodForAll tracks food rescued (kg), CO₂e prevented, night-shift orders served, people fed through NGOs, and progress indices for:
- SDG 2: Zero Hunger. The share of donated food that reached people before its safe time ran out.
- SDG 12.3: Halve Food Waste. The share of collected surplus that was eaten rather than wasted.
⚠️ What's Simulated
- Data: restaurants, accounts, orders and history are sample data generated for demonstration. The forecasting pipeline is real and retrains on actual closing logs.
- Sensors: temperature sensor readings are simulated.
- Weather and calendar: both are real (Open-Meteo and the
holidays package).
🙏 Acknowledgements
Built with 💚 for the night shift
