​
Problem

When adopting Large Language Models (LLMs), businesses face a critical question: Should you self-host the model or simply access it through an API? The right choice depends on several key factors, for example:
  • Do you operate in a regulated industry where owning and protecting proprietary data is critical?
  • How will it impact or augment existing business processes?
  • Do you have the necessary engineering resources and expertise to build, deploy, and maintain an in-house LLM solution from scratch?
Self-hosting LLMs requires advanced DevSecOps and Kubernetes expertise, placing a high operational burden on organisations. The shortage of platform engineering skills raises the barrier to entry, often diverting data teams from integrating LLMs into business systems and delaying real-world impact.

​
Solution

With Kubox, your teams can deploy a self-hosted LLM in minutes — without needing Kubernetes or cloud expertise. This end-to-end example includes full source code to deploy an open-source chatbot solution, bringing the power of Large Language Models (LLMs) into your own secure environment, while giving you full control and significantly reducing operational costs.

​
Data Infrastructure

The system uses Ray.io and vLLM to efficiently serve a Meta-Llama-3.1 model on NVIDIA L4 GPUs instance on AWS.

​
Source Code

Visit our Github repository at https://github.com/kubox-ai/chatbot

​
What’s next

In future examples, we’ll dive into practical challenges and solutions for optimising LLM inference performance to build cost effective, scalable and high-performant AI systems.