AI in Azure

The AI agency for Microsoft Azure

We build AI applications that make it into production. On one platform in your own tenant, with costs per project and clear rules on which data goes to which model.

As an AI agency, we work in Microsoft Azure, inside your tenant, not ours. We tell you which processes are worth it, build the application and run it on a platform you control. That is how a pilot turns into a tool your teams use every day.

to file a customs declaration, fully automated
Seconds instead of days
not needed, even as the workload keeps growing
10+ hires
until incoming mail reaches the right team, down from 12 to 24 hours
< 3 minutes

Shadow AI: your people already use AI. The question is which one.

If the company doesn't offer its own AI, people use whatever is in their browser. That is understandable and usually well meant. For you, it means not knowing which data leaves the building.

This page is for you if …

  • employees paste texts, contracts or customer data into external chat tools because there is nothing comparable in-house.
  • every department buys its own AI licences and nobody knows the total cost.
  • every AI pilot gets its own infrastructure, sometimes even its own tenant.
  • pilots are running, but none of them makes it into everyday work.
  • your data protection officer asks which data goes to which model, and nobody has an answer.

Banning it rarely works. What works is an official route that is easier than the unofficial one.

One platform for all your AI applications

With MODUS, our AI gateway, every AI application runs through a single point in your own tenant. Each application is a project with its own usage and costs. You decide centrally which models a project may use and what gets logged: permanently, in memory only, or just metadata such as tokens.

MODUS is the product version of a platform we built for a life sciences group. Without a central platform, every AI use case quickly gets its own infrastructure, sometimes its own tenant. There, every AI call now goes through one point, and running it costs up to 30% less.

MODUS is not tied to Azure. Running on Kubernetes, it also works on-premises or in another cloud.

MODUS features and pricing

What we build on the platform

The platform is the foundation. The value comes from the applications running on it, and we build those with you, from the first conversation to production. The models usually come from Microsoft Foundry in your tenant, and from other providers where needed.

Answers from your documents (RAG)

Policies, contracts, manuals: the AI finds the right passage and answers with source and document version. The platform controls who may see which document.

Details

Documents are indexed and searched with Azure AI Search (Foundry IQ). Complex questions are broken down into sub-queries and answered in parallel (agentic retrieval). We connect SharePoint, Blob Storage, databases and line-of-business systems. Every answer names the document and version, so it holds up in an audit.

Agents that act (MCP)

Creating tickets, maintaining master data, sending reports: through the Model Context Protocol, the AI works directly in Jira, ServiceNow or SharePoint. Roles and approvals define what it may do, and every action is logged.

Details

Sign-in via OAuth 2.1 as defined in the MCP specification, permissions via roles. Actions with consequences need a human approval and can be reversed.

Data from documents and images

Archives nobody can keep track of anymore become a structured data source. Dashboards, analyses and predictions run on top of it.

Details

The AI reads documents and images, picks out the relevant information and stores it in a structured way. It works its way through in stages: first the folder structure, then the document titles, then a quick pass over the content. Only in the relevant documents does it search for the data and extract it.

What we learned: AI also helps us build

When people think of AI, they think of chatbots or pattern recognition. For us, something else has changed at least as much: how we build software.

For years we kept getting the same request: can you turn this Excel file into a proper application? These were lists and spreadsheets that had grown over years, with entire processes depending on them. Most of the time we had to say no, because the effort was out of all proportion to the benefit.

Today we say yes. With language models, we extract the logic from the file and turn it into code. What never used to pay off now does. The AI isn't an agent in production here, it's our partner in development. The result is a maintainable application your team can develop further, or we can with LUNA.

That is what happened at a life sciences group: the calculations for rebate settlements with health insurers lived in complex Excel files. Today an application checks every settlement automatically against contract and price list. Without AI in development, the project would not have been affordable.

Got a file like that? Bring it to the use case check.

What it has achieved for our clients

Five projects, all in production.

Seconds instead of days

Life sciences group

Automated customs declarations

Every customs declaration used to go out by email, and you had to know which declaration fit which goods. Today the AI creates them fully automatically. Project duration: 77 days.

10+ hires avoided

Life sciences group

Checking inventory write-down claims

Customer service no longer checks every claim in detail, only samples or cases the AI flags. The workload keeps growing, the team doesn't have to.

50 GB of documents turned into data

Life sciences group

Data from unstructured documents

What used to be impossible now works: dashboards run on the data, documents can be improved where it matters, and how authorities will respond can be predicted with a given probability.

Details

About 50 GB of documents sat on different drives, many stored two or three times, some describing processes that run for months. Nobody could keep track of it, let alone analyse it. Language models work through it in stages: first the folder structure, then the document titles, then a quick pass over the content. Only in the relevant documents do they search for the data and extract it.

Live at 3 clients

Website maintenance by AI instead of a CMS

Marketing describes changes in Teams or an interface of their choice, and the AI implements them in Azure. No CMS back end needed.

< 3 minutes instead of 12 to 24 hours

Bank, 1,000 employees

Digital mailroom and SOP assistant

Incoming mail is read, classified and delivered to the right team. The AI answers questions about work instructions with a source.

Details

Incoming mail and questions about work instructions took 12 to 24 hours and slowed the business units down. The solution extracts the text, indexes it with metadata and answers with a source reference, through a secured interface with logging and versioning. Result: delivery in under 3 minutes, measured in production, and more than 800 hours saved per year across 60,000 documents. From PoC to rollout took 6 weeks.

How an AI project with us works

  1. Conversation. Cost check, use case check or architecture call, 60 minutes each. Afterwards you know whether and where AI pays off for you.

  2. Pick the use case. We agree on one process and on what we measure before and after.

  3. Set up the platform. If there isn't one yet, MODUS goes into your tenant. Setup takes about a month.

  4. First use case goes live. For the projects on this page, it took between four and eleven weeks to reach production.

  5. Run and grow. You run the platform yourself, or we take it over with ASTRA, our Azure managed service. New applications are added as further projects.

Frequently asked questions about AI in Azure

What makes an AI agency for Azure different?

We build not just the application but also the platform it runs on, inside your Azure tenant. Costs, permissions and logs stay with you, and the next application doesn't start from scratch.

Does our data stay in our tenant?

MODUS and the applications run in your tenant, and you decide what gets logged. Where a request to a model goes depends on the model and how it is deployed, for example in an EU data zone or with a provider outside Azure. You decide this per project.

Which AI models can we use?

All models in Microsoft Foundry: besides the OpenAI models, that includes Claude from Anthropic, Mistral, Meta and others. MODUS can also connect providers outside Azure.

How fast can a first use case go live?

For the projects on this page, between four and eleven weeks. If MODUS has to be set up first, add about a month for that.

Why MODUS instead of Microsoft's built-in tools?

Microsoft's AI gateway requires Azure API Management, MODUS doesn't. And MODUS doesn't depend on the roadmap of a Microsoft product team. AI moves faster than product teams can ship. If you want an edge, you can't wait for Microsoft to catch up.

Next step

Ready to create impact?

Tell us briefly what it’s about – by email or in a non-binding conversation. We listen, ask the right questions, and show how we can help in a solution-oriented and pragmatic way.

What happens next

  1. 01

    Describe your case

    Three fields, no sign-up. Two minutes is enough.

  2. 02

    Personal reply

    Stefanie Heine gets back to you within one business day.

  3. 03

    Non-binding first conversation

    We listen, ask the right questions, and show how we can help.

Prefer to pick a slot directly?

Your contact

Stefanie Heine

Stefanie Heine

Executive Assistant

0/500 characters

We respond within one business day.

Microsoft Certified Expert Badge50+ customers trust DEVDEERAzure since 2015Microsoft Cloud Solution Provider

Glad to have you here!

To help you quickly find what you’re looking for - or just as quickly realize this might not be the right place - we collect anonymized usage data. Not for advertising, but to make this site work as well as possible for you. Honestly: if we could ask you directly, we would. Thank you for your trust!