Skip to main content

Architecture

Kelvin consists of several interconnected services:

  • Kelvin web
    • Django 4 application that provides a REST API and also a set of user-facing web pages and templates
    • It renders some part of the web using a JavaScript frontend
  • JavaScript frontend
    • Reactive frontend implemented using Svelte 3 and Vue.js, which is rendered into a single JS and CSS file and served by the Django app.
  • PostgreSQL database
    • Provides persistent storage of users, classes, subjects etc. for the web
  • Redis
    • Provides a temporary (non-persistent) cache used for e.g. storing code plagiarism detection check results

The web also uses RQ (Redis) queues for executing various jobs. There are multiple queues:

  • default queue performs jobs related to common asynchronous stuff, like:
    • Code plagiarism detection
    • LLM code submission review
  • evaluator queue performs evaluations of code submitted by students
    • The evaluation logic is implemented in the evaluator directory
    • It uses Docker internally to perform various evaluation pipelines
  • cuda queue also performs evaluation, but it should be used for code that needs an NVIDIA GPU

Here is a simple diagram that describes the architecture:

Directory structure​

kelvin
├── api - REST API of the backend
├── deploy - automatic docker deployment
├── docs - project documentation site
├── evaluator - evaluation pipelines for checking submitted code
│   ├── images - docker images for custom pipeline actions
│   └── pipelines.py - integrated pipeline actions - e.g. Docker evaluation
│── frontend - JavaScript frontend
├── kelvin - base configuration of the application
├── quiz - module for create quizes as tasks
├── survey - module for creating surveys defined in yaml
└── web - web interface for the kelvin web app