🛡️ KitchenGuard AI

Real-Time Computer Vision & Automated Hygiene Compliance for Commercial Kitchens

Live Web App & AI Sandbox • API Swagger Docs

---

KitchenGuard AI is an intelligent, high-accuracy computer vision platform engineered for commercial kitchens, food processing plants, and culinary workstations. Powered by fine-tuned YOLOv11 and an automated Spatial Deduction Engine, it verifies Personal Protective Equipment (PPE) adherence, flags hygiene infractions in real-time, logs every scan to a secure database, and automatically generates official legal citation documents under FDA Food Code & HACCP regulations.

---

✨ Key Features & Capabilities

• 🎯 High-Precision Detection (~90% Precision / 89.2% mAP Gloves): Fine-tuned on specialized food prep datasets for robust, real-world generalization across varying lighting, angles, and skin tones.
• 🧠 Spatial Deduction & Absence Reasoning: Automatically localizes worker facial/head geometry to flag missing PPE (no_hairnet, no_mask) even when zero PPE is worn.
• ⚡ Sub-250ms Low-Latency Inference: Fully containerized and optimized with CPU-tailored PyTorch & OpenCV headless pipelines.
• 📱 Real-Time Telegram Incident Dispatch: Automatic alert broadcast to kitchen managers with timestamped logs, PPE telemetry, and severity status.
• 🎨 Interactive AI Sandbox: Drag-and-drop web dropzone supporting both single-frame photo analysis and full MP4 video clip scanning.
• 🗂️ Secure Audit Database: Every scan is automatically saved — timestamp, inspector, media type, violations, annotated snapshot — with JWT-authenticated user history.
• ⚖️ Legal Citation PDF Generator: Automatically generates formal FDA/HACCP statutory citation documents (Form KG-802) for every logged violation, downloadable directly from the audit table.
• 📜 Regulatory Alignment: Fully mapped to HACCP, FDA Food Code, FDA 21 CFR § 11, and OSHA 29 CFR 1910.134 legal frameworks.

---

🏷️ Detection Taxonomy

Category	Class / Flag	Color Tag	Behavior & Meaning
PPE Compliance	hairnet	🟢 Green	Verified clean hairnet / chef hat covering hair
PPE Compliance	mask	🟢 Green	Verified protective face / beard mask covering mouth & nose
PPE Compliance	gloves	🟢 Green	Verified sanitary food-handling gloves (latex / nitrile)
Hygiene Infraction	no_hairnet	🔴 Red	Worker head detected without required hair covering
Hygiene Infraction	no_mask	🔴 Red	Worker face detected without protective face mask
Hygiene Infraction	no_gloves	🔴 Red	Worker hands detected without food-safe gloves

---

⚖️ Legal Citation PDF System

When a violation is logged, KitchenGuard AI generates a formal regulatory citation document (Form KG-802) that includes:

Document Section	Content
Citation Reference	Unique audit ID (KG-AUDIT-XXXXXX-YYYYMMDD)
Regulatory Statutes	Exact FDA Food Code section numbers & verbatim legal text
Severity Classification	Class I Critical / Class II Severe / Class III Moderate
Corrective Actions	Mandatory steps required by floor supervisor
Penalty Guidelines	Applicable fine ranges and re-inspection windows
Photographic Evidence	Embedded annotated snapshot with AI bounding boxes
Sign-off Block	Supervisor signature & license attestation area

FDA Statute Mappings

Violation Type	Regulatory Code
Missing Hairnet	FDA Food Code § 2-402.11 – Hair Restraints
Missing Mask	FDA Food Code § 2-201.11 + OSHA 29 CFR 1910.134
Missing Gloves	FDA Food Code § 3-301.11 – Bare Hand Contact Prevention

Endpoint: GET /api/scans/{scan_id}/pdf — returns application/pdf with authenticated access control.

---

📊 Model Architecture & Benchmarks

The production system runs the kitchen-hygiene-model checkpoint (YOLOv11s):

Model Summary: 126 layers, 9.4M parameters, 640px resolution
Latency: ~2.4 ms GPU (RTX 4060) / ~180 ms CPU (Cloud Container)

Class	Precision (P)	Recall (R)	mAP@0.50	mAP@0.50:0.95
gloves	87.3%	83.8%	89.2% ⭐	58.1%
mask	85.3%	68.4%	74.5%	35.6%
hairnet	77.4%	57.1%	63.6%	25.6%
Overall Peak	90.08%	79.48%	77.53%	39.30%

---

🚀 Quick Start (Local Setup)

1. Clone & Install

git clone https://github.com/BigSmokeweb/KitchenGuard-A1.git
cd KitchenGuard-A1

# Create virtual environment
python -m venv venv

# Windows
.\venv\Scripts\activate
# Linux / macOS
source venv/bin/activate

# Install dependencies
pip install -r requirements.txt

2. Configure Environment

cp .env.example .env
# Edit .env and fill in your values

JWT_SECRET_KEY=your-secret-key-here
TELEGRAM_BOT_TOKEN=your_bot_token_here   # optional
TELEGRAM_CHAT_ID=your_chat_id_here       # optional

3. Start the Inference Server

python -m uvicorn scripts.server:app --host 0.0.0.0 --port 8000 --reload

Open http://localhost:8000 in your browser.

---

🐳 Docker & Cloud Deployment

Build and run anywhere with the optimized production container:

# Build the Docker image
docker build -t kitchenguard-ai .

# Run locally
docker run -p 8000:8000 --env-file .env kitchenguard-ai

One-Click Railway Deployment

1. Connect your repository to Railway.
2. Railway auto-detects Dockerfile and railway.json.
3. Set environment variables (JWT_SECRET_KEY, TELEGRAM_BOT_TOKEN, TELEGRAM_CHAT_ID) in the Variables tab.

---

📱 Telegram Alert Integration

To receive instant push notifications whenever a kitchen safety violation is detected:

1. Create a bot using @BotFather and copy your TELEGRAM_BOT_TOKEN.
2. Get your Chat ID by messaging @userinfobot.
3. Add to your local .env file (or Railway Dashboard):

   TELEGRAM_BOT_TOKEN=your_bot_token_here
   TELEGRAM_CHAT_ID=your_chat_id_here

4. Test delivery instantly at: http://localhost:8000/test-telegram

---

📄 License & Attribution

Distributed under the MIT License. Created & maintained by Milind Sahu.