How noufmsh/tagie works
Project Type and Purpose
• Tagie is described as an application that connects ML developers with expert human labelers. This indicates it is a service-oriented application (likely full-stack) built around facilitating data labeling tasks that support ML workflows.
• Its purpose is to ensure data is correctly labeled from the start so ML developers can improve products, services, and processes.
Core Architectural Components to Highlight
a. Frontend Components
• User Dashboards for ML developers and separate interfaces for expert labelers.
• A Web Portal or Mobile App that facilitates task submission, review, and feedback.
b. Backend Services
• API Gateway and application servers performing business logic, task management, and coordination between users.
• Microservices or modules for user management (registration, roles, authentication) and for task orchestration (assignment, progress, quality control).
c. Data Storage
• A database service to manage information on users, projects, tasks, and labeling results.
• Possibly a file storage or blob storage component if data sets and images are part of the labeling process.
d. Integration & External Services
• Integration with external tools or services used by ML developers (for instance, ML model management systems or version control).
• Payment or billing modules if the application monetizes the labeling process.
e. Notification/Communication Mechanisms
• Messaging or notification services to alert labelers about new tasks, or notify ML developers about task completions.
Relationships and Data Flow
• Connection between the backend and the database, ensuring that data persistence (e.g., storing label tasks and results) is clearly depicted.
• If your architecture envisages microservices, show how these services communicate with one another (for instance, task assignment services talking to notification services).
Architectural Patterns and Technologies
• Given the dual user base (ML developers and human labelers), you might want to emphasize a modular or layered architecture that cleanly separates concerns – for example, using RESTful APIs or GraphQL APIs between the frontend and backend.
• Possible frameworks or libraries based on what a typical full-stack application might use (like React/Vue/Angular for the frontend and Node.js/Express, Django, or Flask for the backend). Even if not explicitly stated, labeling technology (such as workflow managers or job queues) may also play a role.