Hjem » “ME AI” Voice Assistant

We Built an AI Assistant for Employees Across the Globe

Client
“ME AI” Voice Assistant
Industry
Manufacturing / Enterprise Operations
Challenge
Building one internal AI assistant across Denmark, Mexico, and China
Results

A unified employee assistant for navigation, booking, and internal knowledge

Scope

When a large Nordic-based global leader in the consumer retail market needed a smarter way for employees to navigate headquarters, coordinate daily work, and access relevant information, they set out to build one internal AI assistant that could work across multiple countries.

BCT helped design and build the infrastructure behind it.

The company's headquarter is large, complex, and busy. Employees lose time every day finding their way, locating available meeting rooms, and staying on top of the information that actually matters to their role and location.

The ambition was to create one internal AI assistant that could reduce this everyday friction across the organisation.

The challenge was not just product-related. It was operational and legal. The system had to run across multiple jurisdictions at the same time, each with different rules around AI models, data handling, and compliance.

The scope included:

  • Building one shared assistant experience for employees across the World
  • Supporting everyday tasks such as room booking, wayfinding, and internal communication
  • Creating a voice-first experience that felt natural in use
  • Grounding answers in the company's  own internal knowledge
  • Keeping data inside the company's own infrastructure
  • Adapting model usage and system behaviour to local legal requirements

Concept

A voice-first internal assistant

The result was "ME AI" — a voice-first internal assistant built for the way people actually work inside a large organisation.

Instead of adding another system to learn, the assistant was designed to remove friction in the moments that slow people down.

Employees can book meeting rooms in seconds, navigate the building through a rendered 3D model, call elevators on demand, and receive role- and location-specific news directly through the assistant.

From the beginning, the concept was built around three principles:

  • Useful in daily work: The assistant solves practical tasks employees face every day.
  • Voice-first interaction: Built on OpenAI’s Whisper model, the system allows employees to interact naturally without relying on a keyboard.
  • Automation with control: Critical actions require confirmation before execution, creating a human-in-the-loop safety layer.

Process

Knowledge infrastructure

We built a RAG-based data pipeline that fetches content from SharePoint via API, processes it, converts it into embeddings, and stores it in a vector database. This ensures employees receive answers grounded in real, current company knowledge.

Backend architecture

The backend was built with HONO and Prisma, creating a fast, lightweight, and production-ready setup capable of supporting multiple tools, workflows, and legal configurations.

Frontend experience

The interface was built in React and designed to integrate naturally into the company's existing digital environment. The experience was kept simple, clear, and intuitive to support fast adoption internally.

Infrastructure and deployment

The solution was hosted on Azure and connected through the Vercel AI SDK, while keeping the company's data and model setup within their own infrastructure.

Jurisdiction-aware AI orchestration

One of the most complex parts of the build was engineering the system to adapt automatically by country. Employees in Billund, Monterrey, and Shenzhen all use the same interface, but under the hood, model selection and tool execution shift depending on local legal requirements. This made it possible to deliver one coherent product experience across regions without compromising on compliance.

Results

"ME AI" turned a complex internal challenge into one assistant experience that supports employees across locations and workflows.

The solution made it possible to:

  • Reduce everyday friction in a large and complex workplace
  • Simplify room booking and navigation across a multi-floor headquarters
  • Surface more relevant internal communication based on role and location
  • Enable natural voice interaction from the start
  • Keep sensitive company data inside the company's own infrastructure
  • Deploy one assistant globally while adapting the AI stack to local legal requirements

More broadly, the project proved that enterprise AI can be both useful and compliant — even across regions with very different legal frameworks.

Building something complex? Good.

If you're working on enterprise AI, internal tooling, compliance-sensitive deployments or systems that need to work across regions.

Talk to Us About Your Project.

Danmarks Statistik

A hybrid AI architecture enabled secure search, internal chat, and direct API access

We build scalable software and AI solutions that actually work
– pragmatic, data-driven and made by real people.
Born digital. Built different.
Get in touch
info@blackcapitaltechnology.com +45 60 92 92 60
Mølleå 3-5,
9000 Aalborg
Danmark
(+45) 60 92 92 60
CVR:
DK-42380784
Turbinveien 9,
0195 Oslo
Norway
(+47) 907 00 863
Org. nr.:
NO-933435458