Technology Stack
Python, Flask, AWS (EC2, S3, SQS, Lambda, DynamoDB), YOLOv8, RESTful API, JWT Authentication, OpenStreetMap/Leaflet, HTML, CSS, Docker, ONNX
Project Overview
Designed and implemented a cloud-native AI photo management system that enables intelligent organization of images through automated classification and metadata extraction. The platform supports scalable storage, asynchronous processing, and location-based visualization, providing an efficient solution for managing large photo collections.
Key Features
- Intelligent Image Classification
- Integrated a YOLOv8-based model in ONNX to automatically categorize uploaded images and generate confidence scores.
- Enabled metadata-driven search and organization.
- Asynchronous Processing Pipeline
- Implemented an event-driven architecture using Amazon SQS and AWS Lambda.
- Decoupled user uploads from compute-intensive ML inference to improve scalability and responsiveness.
- Cloud-Native Storage
- Stored image assets in Amazon S3 and managed structured metadata in DynamoDB.
- Designed efficient data models for user accounts, image labels, locations, and processing status.
- RESTful Backend Services
- Developed a Flask-based REST API supporting authentication, image upload, metadata retrieval, and album management.
- Implemented JWT-based authentication and access control.
- Location-Based Visualization
- Extracted EXIF GPS metadata from images and visualized them using OpenStreetMap/Leaflet for map-based browsing.
- User Interaction Enhancement
- Designed a follow-up question mechanism to collect additional metadata, improving search accuracy and organization.
System Architecture
The system follows a modular cloud-native architecture:
- Frontend: HTML/CSS interface for user interaction.
- Backend: Flask REST API deployed on AWS EC2.
- Asynchronous Processing: Amazon SQS and AWS Lambda for background tasks.
- Storage: Amazon S3 for images and DynamoDB for metadata.
- Machine Learning: YOLOv8 for image classification.
- External Services: OpenStreetMap for geospatial visualization.
Technical Contributions
- Built an end-to-end serverless ML pipeline integrating storage, messaging, and compute services.
- Designed scalable NoSQL data schemas optimized for metadata querying.
- Implemented event-driven workflows to enhance system reliability and performance.
- Enabled secure communication using HTTPS and IAM-based access control.