LLMs management portal

LLMs management portal

Last month, one of our teams finished developing an LLM management portal. In this post, I’ll briefly describe this SaaS solution, the time we spent on development, the technologies we used, and provide a link to the development environment.

Our customer had already created an LLM management environment. It allows users to run and train LLMs, feed prompts, and get responses. It is mostly used by educational platforms such as OpenEdx. Owners of such platforms can train LLMs by feeding them books and courses, use them to teach, and test students with quizzes generated by the LLM. All interaction happens through a chatbot that acts as a tutor. In short, this is a sophisticated tool that easily integrates into existing educational platforms.

The client’s request was to develop a platform where their customers would be able to add their own LLMs, use predefined models, run training, and customize bot behavior on educational sites. The platform also needed to support payments and subscriptions. The entire development process—from design to final delivery—had to be completed within two months.

Due to the very limited timeframe, we chose the tech stack we were most familiar with:

  • Backend: jHipster on Java 23 (we usually prefer Kotlin for backend development, but kHipster is not yet up to date with jHipster).
  • Database: PostgreSQL
  • Frontend: Angular web app
  • Payment gateway: Stripe
  • CI/CD: GitHub Actions
  • Deployment: Docker image, runnable via Docker Compose or Kubernetes
  • Ticketing system: Jira
  • Documentation: Confluence
  • Team communication: Slack
  • Customer communication: Telegram

 

In six weeks, we completed the application, and then spent two weeks on integration, testing, and fine-tuning. The team consisted of one backend developer, one frontend developer, one manual QA, and one manager.

You can see the result here: https://portal.intela-bot-ai.com — this is the development environment. Right now, a few production instances are being prepared for educational and other types of customers.

 

We screenshots attached below