Case study
Bringing private AI in-house with PUPAU: installation on dedicated infrastructure
schedule3 min
calendar_today13 February 2026
syncCaricamento...
In recent years Artificial Intelligence has become part of everyday work in many companies. Tools like ChatGPT have made AI accessible to everyone, but there is a fundamental difference between public AI and private AI.
More and more organisations are choosing to bring AI directly inside their own infrastructure, keeping full control over data, security and costs.
In this article we describe a real case: the installation of PUPAU, the platform that orchestrates public and on-premise AI models, on dedicated hardware at the client's site.
Public AI vs private AI: why companies are changing approach
Most AI tools today work through cloud services:
data is sent to external platforms that process the requests and return the results.
This model is perfect for personal use or for non-sensitive activities, but it can become a limitation when it comes to:
- internal company documents
- proprietary data
- business-critical processes
- unpredictable usage-based costs
Private AI turns this model around:
the models run on dedicated infrastructure and the data never leaves the company.
The main advantages:
- full control over data
- greater security and compliance
- predictable costs over the long term
- direct integration with company systems
What PUPAU is
PUPAU is a platform designed to orchestrate AI models in a unified way.
It allows companies to:
- use public models when needed
- run local models on-premise
- manage everything through a single infrastructure
In practice, PUPAU becomes the company's AI gateway.
The project: full deployment at the client's site
Shellonback handled the deployment and the complete installation of the AI infrastructure, from remote preparation through to on-site go-live.
The goal was to create a ready-to-use environment for:
- orchestration of LLM models
- management of the AI backend services
- future integration with company systems
Hardware installed
To ensure high performance and scalability, the infrastructure was built on dedicated machines:
- 1 Ć NVIDIA DGX Spark
- 4 Ć Dell ProMax GB10
This combination makes it possible to handle intensive AI loads and distributed workloads efficiently.
From remote preparation to on-site installation
The project was carried out in several phases.
1. Preparing the deployment
We designed an automated installation system based on:
- Kubernetes (K3s)
- Helm
- Ansible
- full containerisation of the services
This allowed us to prepare the entire infrastructure remotely.
2. Installation at the client's site
Once the machines were ready, we moved on to the on-site phase:
- hardware configuration
- installation of the Kubernetes cluster
- full deployment of PUPAU
- go-live
The result: private AI up and running in the company.
The result
The company now has a ready-to-use AI platform that allows it to:
- use LLM models securely
- keep data inside the infrastructure
- scale its use of AI over time
- integrate new models and services
In other words: AI is no longer just an external tool, it becomes an integral part of the company's infrastructure.
The future of AI is hybrid
More and more companies are adopting a hybrid approach:
- public models when flexibility is needed
- local models when security and control are needed
PUPAU enables exactly this scenario.
Want to bring private AI into your company?
Shellonback supports companies in designing and deploying dedicated AI infrastructure.
If you want to understand how to bring AI inside your organisation, contact us.