Cookies on this site

We would like to use cookies to understand how the site is used and to measure our campaigns. Nothing is set until you choose, and you can change your mind at any time from “Cookie settings” at the foot of every page. Privacy policy

Contact UsContact Us

Build, test and run AI models and agents on Azure AI Foundry.

Azure AI Foundry, which Microsoft now calls Microsoft Foundry, is where custom AI on Azure is built and monitored. We use it to choose and evaluate models on your own documents, connect them to your data, and deploy agents that log everything they do. When a model needs to stay on the device or in the building, we run it locally with Foundry Local or on an NVIDIA desktop AI system.

Talk to our teamSee related work

What we build

Custom AI for the work low-code tools cannot carry

Copilot Studio is a quick way to put an agent in front of Microsoft 365 users, and many of our agents start there. Foundry is for the parts that need more: a model chosen and tested for the task, document processing at volume, and code that runs on Azure rather than inside a low-code flow.

For a national food cooperative we built a label comparison agent in Copilot Studio, then moved its document processing to Azure Functions, with Azure AI Document Intelligence as part of it, when the work outgrew the low-code platform. It compares two versions of a label and returns an audit-ready list of differences in one chat turn.

What we do

  • Model selection and evaluation on your data
  • Document Intelligence for extraction and comparison
  • Agents connected to SharePoint and your business systems
  • Agent deployment and monitoring

Approach

One workflow, measured, before anything is scaled

Every engagement starts with one bounded job and a clear measure of success. We test candidate models on your real documents and pick the smallest one that passes your accuracy test, because it is usually the fastest and the cheapest to run. A pilot follows with a reviewer in the loop, then production with least-privilege access, managed identity instead of stored passwords, and full activity logs. After launch we keep tuning as your documents and processes change.

What we do

  • Discovery sprint: one job and a success measure
  • Pilot on your real documents with a reviewer in the loop
  • Production with least-privilege access and activity logs
  • Ongoing tuning

Governance

Every prompt screened, every action logged

Each agent we build ships with a safety gateway and a log of everything it did. Purview sensitivity labels travel with the data, Entra ID decides who and what can reach it, and Princeton Sentinel, our own product, runs inside your Azure tenant to log the prompts and tool calls of your Copilot and Copilot Studio agents, score each agent's risk and screen prompts for injection attempts.

AI on the edge

When the model should not leave the device or the building

Some work should never reach a cloud model: sensitive documents, sites without a reliable connection, or volumes where per-call pricing adds up. Foundry Local runs a model on the laptop or server itself, with no network round-trip. For heavier batch work, a desktop AI system on the NVIDIA GB10 chip, the chip in NVIDIA DGX Spark, runs larger models next to the team that uses them.

Our own bid-triage pipeline runs a 30B model on one GB10 desktop in our office. On the same bids it was 11 times faster than a 120B model with the same verdicts, and no document text is sent to a cloud model. Our Hybrid & Edge AI page, under AI Consulting, shows how we decide which tier each task runs on.

What we do

  • Foundry Local for on-device AI
  • NVIDIA DGX Spark and GB10 desktop systems for departmental batch work
  • Routing rules that send each task to the right tier
  • One activity log across device, your servers and the cloud

Case studies

Work on Azure AI Foundry

Client names are withheld. We are happy to share details on a call.

All case studies
  • Delivered

    Every change, flagged

    Food manufacturing

    AI Consulting

    Label comparison agent for a national food cooperative

    Compares two versions of a product label, highlights every change, annotates the PDF and returns an audit-ready list of differences. We moved it from Copilot Studio to Azure Functions when the document processing outgrew the low-code platform.

    • Audit-ready differences in one chat turn
    • Rebuilt from low-code to cloud functions
  • Internal project

    2,000 → 50 a day

    Public-sector sales

    AI Consulting

    Procurement triage on a local 30B model

    Logs in to five government procurement portals each morning, reads about 2,000 listings, parses the bid documents and emails a short list of the 20 to 50 worth quoting. Every model call runs on one desktop AI system in our office.

    • 2,000 listings down to about 50 a day
    • 11× faster than a 120B model, same verdicts
  • Product

    Every prompt screened

    Any industry on Microsoft 365

    Microsoft

    AI activity and data posture for Microsoft 365

    Sentinel watches Copilot and Copilot Studio agents in a Microsoft 365 tenant. It logs prompts and tool calls, scores each agent's risk, flags oversharing and stale access, and screens prompts for injection attempts with a classifier that runs locally at no per-call cost.

    • Every agent prompt screened
    • Runs inside the customer's own Azure tenant

Questions

Common questions

Is Azure AI Foundry the same as Microsoft Foundry?

Yes. Microsoft renamed Azure AI Foundry to Microsoft Foundry in late 2025, and the Azure AI services inside it became Foundry Tools. It is the same platform, and both names still appear in Microsoft's documentation.

We already use Copilot Studio. Do we need Foundry as well?

Not always. Copilot Studio is the right place for many agents. Foundry earns its place when a task needs a specific model, document processing at volume or custom code, as it did for the label comparison agent we moved to Azure Functions and Document Intelligence.

Which model will you use?

The smallest one that passes your accuracy test on your own documents. Foundry offers models from OpenAI, Microsoft and other providers; we test the candidates on the same inputs and show you the results before recommending one.

Can we use AI without sending our documents to the cloud?

Yes. Foundry Local runs models on the device, and an NVIDIA GB10 desktop system runs larger ones in your office. Our bid-triage pipeline does all of its model work on one such desktop.

How do we start?

With one workflow. Bring the documents it handles and the result you need, and we will test the models on them, run a pilot with a reviewer, and tell you what production would take.

Have a workflow in mind for AI on Azure?

Bring one workflow and the documents it handles. We will test the models on them and show you what it would take to run in production.

Talk to our team