Skip to content

Our AI infrastructure and model-agnostic architecture

We build your processes on the most widely used AI platforms. Which model goes into a project depends on the job, the cost and where your data needs to stay, never on how popular a brand is.

Last updated:

The infrastructure we use

We have more than one option at every layer. Your needs for performance, cost and data security decide which goes into your project.

  • Cognitive power (large language models)

    Understanding, summarising and classifying text; writing content and code

    • OpenAI GPT models
    • Google Gemini
    • Google Vertex AI
    • Anthropic Claude
  • Enterprise cloud and security

    Scalable infrastructure under enterprise contracts, with a choice of region

    • Microsoft Azure AI
    • Amazon Web Services (AWS) AI
  • Generative AI (creative)

    Content made with images, video and digital avatars

    • Midjourney
    • DALL·E
    • Stable Diffusion
    • Runway
    • HeyGen
  • Openness and flexibility

    Open-source models on your own servers, so data never leaves the company

    • Hugging Face
    • Meta Llama (on-premises)
  • Workflow automation

    Cross-application triggers and automation scenarios

    • Make (Integromat)
    • Zapier

The brands and products named are trademarks of their respective owners. dipro.uk uses these platforms in client projects.

How does a model-agnostic architecture work?

We put our own orchestration layer between your data and the AI models, so you never have to rebuild your systems to change a model.

Your data

CRM and ERPCustomers, orders, stock
Email and documentsQuotes, invoices, contracts
Web and e-commerceForms, baskets, behaviour

dipro orchestration layer

  • Business rules and prompt management
  • Access control and audit logs
  • Model choice by task, with cost control
  • Human approval at critical steps

Models

Cloud LLMsOpenAI, Gemini, Claude
Enterprise cloudAzure AI, AWS
On-premises modelsLlama, Hugging Face

How do we place AI in your sector?

Every sector has its own data and its own bottlenecks. Here are sample use cases from four of them.

A hotel lobby.
Tourism and travel

Multilingual booking and smart pricing

Booking assistants that work around the clock in several languages, dynamic pricing driven by historical data and market trends, and travel itineraries personalised in real time.

Staff working in a logistics warehouse.
E-commerce and digital marketing

Personal recommendations and content production

Marketing flows that recommend the right products from customer habits, production lines for SEO content and images, and AI layers that keep improving ad campaigns with data.

An automated production line.
Manufacturing and industry

Supply chain and internal knowledge assistant

AI-optimised supply chain processes, operational efficiency modelling and internal assistants that search your company knowledge base in seconds.

Screens showing financial data.
Finance and data management

Automated reporting and decision support

Automatic processing of financial statements, portfolio movements and market data, risk analysis, and decision-support dashboards for senior management.

Questions about the infrastructure

What does a model-agnostic AI architecture mean?

It means placing an orchestration layer between your systems and the AI models. Business rules, prompts and permissions live in that layer, so changing a model does not mean rebuilding your systems.

Can we use AI without our data leaving the company?

Yes. For sensitive data we can run open-source models such as Meta Llama on your own servers using the Hugging Face ecosystem. In that case the data never leaves the company.

How do you choose a model for each job?

The job, the cost and where the data has to stay decide it. We use large language models for reading and writing text, Azure AI or AWS when data must stay in an enterprise cloud, and generative tools for images and video.

Let’s choose the right architecture for your sector.

Tell us about the systems you use and how sensitive your data is, and we’ll send you a written recommendation for models and infrastructure.