Practical AI, built for your organisation.

I help your organisation automate recurring work with AI assistants for documents, e-mail and business data. From an initial trial to a working solution, on your own infrastructure or in the cloud.

I also help design and develop your Microsoft Fabric data platform.

Arrange an introductory callExplore my work
In three stages from your sources to usable results Stage 1, sources: e-mail, documents, knowledge base, business data, CRM and internal APIs. Stage 2, processing: the data reaches AI agents, locally or in the cloud. Stage 3, results: reports, summaries, notifications and follow-up actions. Your team stays in control.01SOURCES02PROCESSING03RESULTE-mailDocumentsKnowledge baseBusiness dataCRMInternal APIsAI AGENTSlocally or in the cloudYour team stays in controlReportsSummariesNotificationsFollow-up actions
From your sources, through AI agents, to usable results Business APIs, documents and CRM reach AI agents, locally or in the cloud. The results: reports, notifications and next steps. Your team stays in control.01SOURCES03RESULTBusiness APIsDocumentsCRMAI AGENTSlocally or in the cloudYour team stays in controlReportsNotificationsNext steps
What I offer

What I can do for your organisation.

  • 01

    Advice & exploration

    Discover where AI genuinely saves your team time.

    We map your processes and pick the opportunities with the most impact.

  • 02

    Custom AI

    Assistants (agents) that work alongside what your team already does.

    From e-mail and documents to reports, tailored to how you work.

  • 03

    Privacy & control

    Make smart use of AI without handing over your data.

    You decide what an LLM gets to see and what stays inside your organisation.

  • 04

    Process automation

    Let agents handle the recurring work.

    They run your business processes; your team stays in control and approves where it matters.

  • 05

    Local or cloud

    The power of AI, where it suits your data.

    Sensitive data on your own hardware, the rest in the cloud where possible.

  • 06

    Microsoft Fabric & dbt

    A data platform your organisation can build on.

    Architecture, setup and controlled deployment with dbt and Azure DevOps.

    Learn more
Examples

My work in practice.

From speeding up everyday processes with specialised custom agents to setting up and optimising local AI.

In practice · Province of Gelderland

Agents in Microsoft Fabric

From a plain-language request to a change the team can review.

At the Province of Gelderland, I helped set up Microsoft Fabric and its integration with Azure DevOps. As a BI engineer, I also built a platform in which AI agents prepare changes to data models and reports.

The result is a proposed change the team can review: a pull request. The team stays in control: a person approves the plan first and ultimately decides whether to merge and release it.

The agent team
PLANCoordinator
BUILDdbt & Power BI specialists
CHECKReviewer
RECOVERDiagnostics
  • A person approves every plan first
  • Own branch and workspace per change
  • Every step recorded and traceable
  • No automatic merge or release
View all my work
  • Set up and configured a scalable Microsoft Fabric environment for data engineering and analytics.
  • Integrated Fabric with Azure DevOps for Git, version control, deployment and development automation.
  • Designed and implemented automated processes that create isolated Fabric development environments, allowing teams to work independently and with control.
  • Shared responsibility for migrating an existing data warehouse to Fabric, including modernising existing data solutions.
  • Contributed to CI/CD processes, development standards and platform operations.
  • Worked on integration, dbt transformations and BI access alongside data engineers, BI engineers, architects and other stakeholders.

Technologies: Microsoft Fabric · Azure DevOps · Azure · dbt · CI/CD · Git · Data Warehousing · BI

  • Developed AI solutions to further automate development, operations and data engineering processes.
  • Built AI-assisted tools and dashboards that give developers and BI engineers insight into ongoing work and development processes.
  • Built a coordinator that investigates requests in Azure DevOps and Fabric and assesses feasibility. After approval, ingestion, transformation and reporting specialists work in their own feature branches and workspaces, through the existing pipelines.
  • Built a workflow where a change can be requested in plain language. Agents use the current sprint and backlog to analyse the request and carry out the change.
  • Applied AI-assisted development in existing development and DevOps processes to automate repetitive tasks.
  • An independent reviewer checks every change against the approved plan; a diagnostics agent investigates errors. Agents use sprint information, development standards and existing solutions. Merging and production releases remain human decisions.

Technologies: AI agents · Agent orchestration · Generative AI · Azure DevOps · CI/CD

Own lab · NVIDIA Thor & Spark

Local AI on dedicated hardware

Measure first to find the models and hardware that suit your application.

In my own lab, I test local language models on NVIDIA Jetson AGX Thor and DGX Spark. I measure speed and power consumption and explore how multiple machines could run larger models together.

Those measurements and hands-on experience inform advice on models, hardware and local AI within your organisation. One possible application is a shared coding assistant for a small development team, without external AI requests. We test quality, speed with concurrent users and cost against subscriptions in a trial first.

On the workbench
NVIDIAJetson AGX Thor
NVIDIADGX Spark
  • Install models
  • Measure speed & power
  • Explore clusters of multiple Sparks
  • Inform hardware choices
View all my work
  • Installed, tested and optimised local language models on my own NVIDIA hardware, focusing on performance, efficiency and privacy.
  • Set up a personal AI lab to benchmark models, inference techniques and AI architectures.
  • Optimised local AI workloads for speed and cost, with as little reliance as possible on external cloud and API services.
  • Explored and tested model serving, quantization, memory optimisation and GPU utilisation.
  • Tested multi-agent orchestration where specialised agents work together on different tasks.
  • Explored connecting multiple DGX Sparks to support larger models, shared computing capacity and simultaneous use by several employees.
  • Explored architectures for distributed inference, including task distribution, agent communication and using multiple models and machines.

Technologies: Local LLMs · NVIDIA GPU Computing · Python · Model Serving · Quantization · Distributed AI

Own product · SafePrompt

Cloud AI with a privacy filter

Replace detected personal data before a request goes to the cloud.

Texts sent to cloud models can contain names, addresses or dates of birth. My product SafePrompt filters locally: it replaces detected personal data with fictitious values before a request goes out.

This can reduce the amount of personal data in cloud requests. No filter catches everything. We therefore test detection against your data and decide when additional review or local processing is needed.

How it works
INPUTYour request
LOCALPrivacy filter
EXTERNALCloud model
BACKAnswer
  • Detects names, addresses and dates of birth
  • Replaces detected data with fictitious values
  • Runs locally, inside your organisation
  • Test detection against your data first
About me

One point of contact, from advice to rollout.

I’m Mike Modvili. Over twelve years in software and data development, including at the Province of Gelderland, I have built solutions for complex workflows. For the past two years I have specialised in AI workflows and automating tasks with collaborating agents.

My speciality is local AI: language models on your own infrastructure. My lab helps me ground decisions about models and hardware in practical testing. I also work as a Microsoft Fabric engineer on the architecture, setup and development of data platforms.

You work directly with me, from the first question to handover. I provide a working solution, clear documentation and knowledge transfer so your team can continue independently.

Get in touch

Half an hour, no strings attached.

Advice, custom agents, local AI or Microsoft Fabric: tell me briefly what you need help with.

I will reply to your message within one business day to arrange a no-obligation 30-minute conversation.

Prefer e-mail? mike@modvili.nl

How we work together

  1. Discuss your needs

    We discuss your situation and what you want to achieve.

  2. Define the first step

    We agree on the scope and cost of the first step. For local AI, a trial lets us assess quality, speed and usability.

  3. If we continue: build and hand over

    I build and integrate the solution, with documentation and knowledge transfer.