Pawdentify – AI Breed Classifier & Pet Management Platform

Pawdentify is a production-grade full-stack platform that identifies 120+ dog breeds with 89% accuracy and offers a comprehensive suite for pet health management. It bridges the gap between Deep Learning and user-centric design, providing a localized experience in 4 languages.

---

Live Deployment

• Production Frontend: https://pawdentify-frontend.vercel.app
• Production API: https://pawdentify-backend.duckdns.org
• Interactive API Docs: https://pawdentify-backend.duckdns.org/docs

---

System Architecture

graph TD
    subgraph Client_Side [Frontend - Vercel]
        User((User)) -->|Interacts| UI[React UI]
        UI -->|Direct API Call| Maps[MapMyIndia SDK]
    end

    subgraph Server_Side [Backend - AWS EC2]
        Nginx[Nginx Reverse Proxy] -->|Port 443 to 8000| FastAPI[FastAPI App]
        
        subgraph Persistent_Process [Tmux Session]
            FastAPI
            Init[Startup: Load Model] -.->|Once| S3[(AWS S3 - .keras)]
            FastAPI --- LoadedModel[[Resident EfficientNetV2B2]]
        end
    end

    subgraph External_Services [Cloud Services]
        FastAPI -->|Auth| Clerk[Clerk Auth]
        FastAPI -->|Data| Mongo[(MongoDB Atlas)]
        FastAPI -->|Media| Cloudinary[Cloudinary]
    end

    UI -->|HTTPS Requests| Nginx

Loading

Features & Highlights

AI-Powered Identification

• EfficientNetV2B2 Backbone: Utilizes a state-of-the-art EfficientNetV2B2 architecture for high-precision breed classification (Inference time < 3s).
• Deep Breed Insights: Access comprehensive data on temperament, physical traits, and care requirements for 120+ breeds.

Pet Management System

• Digital Pet Profiles: Add and manage multiple pets with customized profiles.
• Health & Activity Tracking: Add persistent notes with categorized tags:
  • Vaccination: Track immunization dates and history.
  • Nutrition: Log food preferences and dietary requirements.
  • Medical: Keep records of medications and vet visits.
  • Activity: Monitor exercise and training progress.

Localized Experience

• Multilingual Support: Fully localized UI in English, Hindi, Urdu, and French via i18next.
• Veterinary Locator: Real-time clinic discovery integrated with MapMyIndia SDK.

---

Engineering & DevOps

• Model Decoupling: The .keras model is decoupled from the application logic, hosted on AWS S3, and fetched dynamically via boto3 on server initialization.
• Production Networking: Secured via Let's Encrypt (SSL) and managed with an Nginx reverse proxy on an AWS EC2 Ubuntu instance.
• Identity Management: Secure user authentication and dashboard synchronization provided by Clerk.

---

UI & Features Gallery

Home & Landing	AI Breed Prediction

Breed Identification Results	Detailed Breed Insights

Veterinary Locator (Map)	Breed Search & Ranking

Pet Profiles & Management	Health Tracking & Notes

---

Technology Stack

Frontend: React 18, Vite, Tailwind CSS, Framer Motion, i18next (Localization).
Backend: FastAPI (Python 3.10), TensorFlow, Cloudinary (Image Hosting), MongoDB.
Infrastructure: AWS (EC2, S3, IAM), Vercel, Nginx, Certbot (SSL), DuckDNS.

---

Engineering Challenges & Solutions

• Model Optimization: Transitioned from EfficientNetB0 to EfficientNetV2B2, achieving a better balance between accuracy (89%) and inference speed for real-time mobile usage.
• Cold Start Latency: Implemented an asynchronous model-loading strategy from AWS S3 on startup to ensure the API is ready for requests immediately upon container health checks.
• Secure Multi-Tenancy: Leveraged Clerk JWTs to scope MongoDB queries, ensuring that pet health notes and history remain private to the authenticated owner.
• Infrastructure as Code: Documented the full deployment pipeline from GitHub Actions to AWS EC2, utilizing Nginx for request buffering and SSL termination.

---

Project Statistics

• Dataset: Stanford Dogs (20,000+ images)
• Model Backbone: EfficientNetV2B2 (Transfer Learning)
• Top-1 Accuracy: 89%
• Average API Latency: 2.4s
• Localization: 4 Languages (i18next)

---

Minimalist Setup (For Development)

Since this project relies on specific Cloud Infrastructure (AWS, Clerk, MapMyIndia), it is optimized for production. To run a local instance:

1. Clone the repo.
2. Provide your own .env file based on the keys listed in the Deployment Documentation.
3. Run the backend via uvicorn and frontend via npm run dev.

Note: For a full walkthrough of the infrastructure setup, please see the Pawdentify – Deployment & Infrastructure Documentation.pdf in the root directory.

---

Built with ❤️ by Devashish Mishra