ENNL
Book a call

AI engineering

What AI can do for your organisation.

We start from one concrete process and a governed platform, and we measure the result on your own data.

Book a callFind your AI starting point

01 / Where AI pays off

Start with one process that pays off

  • ConcreteOne document type, one question or one asset class, so the result can be compared with the current process.
  • GovernedAccess, lineage and audit evidence follow the same rules as the rest of your data.
  • EvidencedAccuracy and time saved are measured on your own data before production.
  • Yours to runThe pipeline is code in your repositories and your Azure tenant, with a run book.

AI typically pays off when it takes over one repeatable piece of work and is measured against today. We pick that piece with you.

Many organisations hear about AI and cannot see where it pays off. A core team and a network of dedicated engineers build it, including the AI harnesses and frameworks that keep it reliable in production.

We build the chosen piece on a governed platform and keep a person in the loop where it matters.

02 / Use cases

Six realistic places to start

From delivered work or a track we hold credentials in. Each use case shows the problem, how we build it and the outcome we aim for.

  • Document automationInvoices, CMRs, orders and customs documents are typed over by hand, which is slow and error-prone. A pipeline reads each document, turns it into structured data and sends uncertain fields to a person for a check, with an audit trail.
  • Self-service data onboardingBusiness users wait on engineers every time a new data source must be added to the platform. A configuration platform lets data owners request a source. After approval, Terraform and Asset Bundle deployments provision it automatically.
  • LLM operations under governanceYou want LLMs on sensitive data, but classification, sovereignty and audit expectations are strict. LLM workloads run on a platform with Unity Catalog governance, so access, lineage and audit evidence follow the rules of your other data.
  • Predictive maintenance from telemetryField assets send high-frequency sensor data, but failures are still found too late to plan maintenance. Streaming pipelines on Azure Event Hubs and Databricks score telemetry as it arrives and flag the assets that need attention.
  • Agent evaluationAI agents are easy to demo and hard to trust, because nobody can say whether a change made them better or worse. We build an evaluation set from your real tasks and run it on every change, so quality is measured instead of argued.
  • An AI harness for your teamsEach team builds its own AI prototype with its own plumbing, and none of them has a way to measure quality or to run reliably in production. We set up a shared harness: agent orchestration and tool access, an evaluation suite with a regression gate, guardrails, tracing and cost control. Teams then add new use cases on the same frame.

Uncertain fields go to a person for a check, with an audit trail.

Document automationIllustration

Extracted fields

  • SupplierDemo supplier
  • Document numberINV-10428
  • Date9 Nov 2026
  • VAT€184.14
  • Total€1,061.00
  • ReferencePO-4471Check

Problem, approach and expected outcome for each use case

01

Document automation

The problem
Invoices, CMRs, orders and customs documents are typed over by hand, which is slow and error-prone.
What an AI system does
A pipeline reads each document, turns it into structured data and sends uncertain fields to a person for a check, with an audit trail.
Expected outcome
Aims to cut manual typing and turnaround time so your team can focus on exceptions. In the DHL RMA Logistics case, invoice processing went from 2 days of manual work to under 10 seconds.
How we deliver
Built in the DHL RMA Logistics engagement. We measure accuracy on your own document types.

02

Self-service data onboarding

The problem
Business users wait on engineers every time a new data source must be added to the platform.
What an AI system does
A configuration platform lets data owners request a source. After approval, Terraform and Asset Bundle deployments provision it automatically.
Expected outcome
Aims to give AI use cases a governed, quickly extended data intake. The RIVM case study reports onboarding time reduced from days to hours.
How we deliver
Built at RIVM. It is not an AI model itself: it is the governed intake that AI use cases build on.

03

LLM operations under governance

The problem
You want LLMs on sensitive data, but classification, sovereignty and audit expectations are strict.
What an AI system does
LLM workloads run on a platform with Unity Catalog governance, so access, lineage and audit evidence follow the rules of your other data.
Expected outcome
Aims to let you use LLMs with the evidence an auditor asks for.
How we deliver
LLM operations are part of our RIVM engagement. We build it on your data and your rules.

04

Predictive maintenance from telemetry

The problem
Field assets send high-frequency sensor data, but failures are still found too late to plan maintenance.
What an AI system does
Streaming pipelines on Azure Event Hubs and Databricks score telemetry as it arrives and flag the assets that need attention.
Expected outcome
Aims to move maintenance from reactive to planned and to balance streaming cost against the latency you need. The energy grid case study reports unplanned outages reduced by 31% in the first year.
How we deliver
Built for an energy grid operator. We start with one asset class and measure the gains on your assets and data.

05

Agent evaluation

The problem
AI agents are easy to demo and hard to trust, because nobody can say whether a change made them better or worse.
What an AI system does
We build an evaluation set from your real tasks and run it on every change, so quality is measured instead of argued.
Expected outcome
Aims to give your team evidence before an agent reaches users.
How we deliver
We completed the Databricks Delivery Expert training track for Agent Evaluation. We build each evaluation set on your own tasks.

06

An AI harness for your teams

The problem
Each team builds its own AI prototype with its own plumbing, and none of them has a way to measure quality or to run reliably in production.
What an AI system does
We set up a shared harness: agent orchestration and tool access, an evaluation suite with a regression gate, guardrails, tracing and cost control. Teams then add new use cases on the same frame.
Expected outcome
Aims to move new AI use cases from prototype to production faster, with quality measured on every change.
How we deliver
We set up such harnesses for clients. We start with one use case and shape the harness around how your teams work. We completed the Databricks Delivery Expert training track for Agent Evaluation.

All AI engineering solutions AI engineering

Voice agents

Voice agents that answer the phone

Built with LiveKit for real-time audio and ElevenLabs for speech. We have built many of them.

  • Answer and route calls to the right team or person
  • Intake and booking, with the details written to your systems
  • Multilingual, including Dutch, with a handover to a person

Problem, approach and expected outcome

VA

Voice agents for inbound calls

The problem
Phones ring out of hours or queue at peak, and staff spend time on routine questions and on passing callers to the right person.
What a voice agent does
An agent answers the call, understands the request, routes it to the right team, takes intake details and books appointments. It speaks several languages, including Dutch, and hands over to a person when it is unsure.
Expected outcome
Aims to shorten waiting time and take routine calls off your team. These are expected outcomes, not results: call volume, languages and integrations decide what is realistic, which is why we pilot on one real call flow.
How we deliver
We have built many voice agents with LiveKit and ElevenLabs. We measure how well it works on your own call flow.

Where call content is sensitive, we can run the language model on infrastructure you control. ElevenLabs is a hosted service, so a fully self-hosted voice agent swaps in open speech models, after we have evaluated whether they are good enough for your languages.

Sovereign and self-hosted AI Talk to the demo voice agent

Inbound callIllustration
Live
  • Hello, I would like to book an appointment.
  • Of course. Which day suits you?

Routed toBooking

Your AI, on your terms

Your AI, on your terms

Some data cannot go to an external AI service. We build self-hosted AI on open-weight models, fine-tuned on your data and run where you decide.

  1. Your dataDocuments, calls, records
  2. Fine-tuneIn an environment you control
  3. Open-weight modelYours, tested on your tasks

  4. You choose where it runs

    1. Nebius, European regions
    2. Your own servers
    3. Azure, EU region

Why it matters

For European organisations that care about sovereignty and owning their data.

  • Your data stays put. Fine-tuning and inference run in an environment you control, so documents and calls do not go to an external AI API.
  • Your model, evaluated. An open-weight model that you own, tested on your tasks and compared with a hosted one before you commit.
  • Your choice of host. Nebius in European regions, your own servers or Azure in an EU region, handed over so your team runs it.

Explore sovereign and self-hosted AI

Governance for AI

One gateway for every model, agent and tool

Every call from teams, agents and coding tools passes through Unity AI Gateway, so access, spend and logs follow the rules of the rest of your data.

  1. Teams, agents, coding toolsWhoever calls a model
  2. Unity AI GatewayPermissions, budgets, limits, policies, logging
  3. Models, MCP servers, toolsDatabricks-hosted and external
  • AccessUnity Catalog privileges decide who may use what.
  • Budgets and limitsSpend and request rates have a cap per use case.
  • PoliciesRules apply to requests and responses.
  • LoggingEvery call lands in Delta tables for audit.
  • RoutingTraffic splits across models, with failover.

How we set it up

  • The gateway and its policies are defined as code and deployed by pipeline, never clicked together.
  • One service per use case, each with its own budget and an owner who answers for it.
  • Inference tables with a retention period and access groups, so logs are useful and not a liability.
  • System-table dashboards for tokens, cost and latency per team.
  • Coding agents such as Claude Code are routed through the gateway, with repository rules and review gates on top.
  • Fallbacks are tested before go-live, by switching a destination off on purpose.
  • Smart Routing stays out of production while it is in Beta.

Generally available. Smart Routing is in Beta. Last verified: 10 November 2026. Unity AI Gateway documentation

AI agents

Agents you can measure and afford

We engineer agents with tools, memory, evaluation and a budget, on the model you choose, and we measure and reduce cost per task without losing quality.

AI agents in depth Cost and token optimisation

03 / A first engagement

From quick scan to production

  1. Data & AI quick scan

    2 to 3 days

    Map data and document flows, rank three use cases.

  2. Document automation pilot

    3 to 4 weeks

    One document type, a human check and an accuracy and time-saved report.

  3. Production

    If the pilot shows value

    Reviewed pipelines, a run book and an agreed operator.

Each step ends in something you can judge. You can stop after any of them.

04 / What you need

What needs to be in place

  • Data you can reachOne document stream or dataset is enough, and someone who knows good output.
  • A named ownerOne person who decides what is correct and reviews exceptions.
  • GovernanceAgreed rules for sensitive data, model location and who sees results. We follow our delivery standards.
  • A platform, as you scaleUse cases that combine systems need a governed data platform.

A first project needs little of this. A second needs more.

Why a modern data platform pays off

05 / Start here

The Data & AI quick scan

Typical duration: 2 to 3 days

Data & AI quick scan

In 2 to 3 days we map how data and documents flow today and rank the three AI or data use cases with the best value for effort.

  • Current state of systems, data and manual work
  • Three use cases ranked by value, effort and risk
  • Roadmap with a pilot proposal
  • Session with management to discuss the result

The usual first step for an AI engagement. Planning and quote follow a first conversation.

Ask for a quote

Each starts with a conversation about your scope.

Academy

Learn AI engineering, from scratch to professional.

academy.nexops.nl has learning paths with the best existing courses, plus projects and checklists. For live sessions, we give workshops to companies and students.

Open the academy Workshops

Free check

Take the check

A score from 0 to 100 and a next step, in about three minutes.

All six checks

06 / FAQ

Questions about AI projects

Is our data safe, and where do the models run?

The document automation pilot runs in your own Azure tenant. We apply privacy by design: only the data a use case needs, EU regions for EU clients, and no production personal data in dev or test. We do not send client data to external AI services without your written agreement.

Do we need a data platform first?

Not for a first pilot on one document type. Use cases that combine several systems need a governed platform, and the quick scan tells you which case you are in.

How long does a first project take?

The quick scan takes 2 to 3 days and a document automation pilot takes 3 to 4 weeks. Production timing depends on the pilot result.

What if the use case does not pay off?

The quick scan ranks use cases by value, effort and risk, so you can decide not to proceed. The pilot ends with an accuracy and time-saved report, so you decide on evidence.

Do you promise a percentage saving?

No. We describe expected outcomes and quote results only from delivered case studies. Your result depends on your data and process, which is why we measure in a pilot.

Next step

Wondering where AI fits?

Tell us about one process you want to improve. We will say whether AI is the right tool, and what a first engagement would involve.

Discuss your use caseSend a messageor email info@nexops.nl directly

Currently accepting projects for Q4 2026