Skip to content
Faissal Makrini

IT Project Manager · AI & Process Automation · Essen, Germany

AI that runs in production. Projects that stay in control.

More than 5 years of end-to-end delivery for IT and software projects: shared responsibility for development budgets up to €15M, teams across Europe and Japan, a backlog cut in half. Since 11/2025 I have been building and running AI process chains myself, measured and safeguarded. This site and its assistant run on my own infrastructure.

See the AI projects

Available immediately · Remote from Essen, Germany: EU and worldwide (CET) · open to on-site workshops

Portfolio · AI projects

Not experiments. Put into production.

Reference projects with examples, measured outcomes and documentation. Each with its starting point, the decision and the result, the way I would write it in a status report.

Automation cockpit · in production since 11/2025

End-to-end AI process chain

Starting pointRecurring research and document work, done by hand every time and slightly differently every time.

SolutionA chain of n8n, language models, REST APIs and webhooks on self-hosted Docker infrastructure: data intake from 6 interfaces, pre-filtering, AI scoring, research and personalised documents.

  • around 80 records per run, approx. 40 % qualified automatically
  • a complete document package in about 3 minutes
  • runs every 20 minutes, unattended

Quality assurance

Measure, don't trust

Starting pointThe language model rated the quality of its own output at 92 % across the board.

DecisionMeasured instead of trusted: actual requirement coverage ranged from 50 to 100 %. That became a hard limit in code with targeted automatic correction.

  • measurable rules enforced in code, not requested from the model
  • a block on invented content; anything unproven is flagged as a gap
  • automated review workflows check every result before release
  • more than 140 automated tests before every release

Vendor and technology decision

Model choice based on data

Starting pointAn analysis of around 1,000 runs showed that one provider rejected every request because of a plan limit, while the fallback quietly absorbed it.

DecisionRoot cause proven via the rate-limit headers, then the model re-chosen per task and tested on 27 real cases against the previous setup.

  • free EU model for volume, a stronger model only where quality drops
  • criteria: quality, data protection with training opt-out, cost
  • automatic failover on outages stays in place

Capstone project of the training

AI in the PMO

Starting pointRisk registers, multi-project overviews and status reports take PMO time that is missing for decisions.

SolutionGDPR-compliant AI agents for risk management, multi-project management and status reporting.

  • overall workflow of six sub-workflows, documented and signed off
  • handover artefacts: short docs, runbook, checklists, UAT
  • human-in-the-loop: approvals stay with people, GDPR from day one

Live on this site

This site and its assistant

SolutionNode back end with a strict content security policy, rate limiting and a server-side token. The assistant runs through a secured n8n workflow; no key ever reaches the browser.

  • labelled as AI under Art. 50 of the EU AI Act
  • no third-party sources, no tracking, self-hosted fonts

Operations, not a demo

Own platform

SolutionCoolify, n8n, Forgejo, NocoDB and a self-hosted search instance behind Traefik, all on my own hardware.

  • versioned, documented deployments
  • certificates, backups and health checks
  • monitoring with push alerts
  • duplicate detection and failover paths
Diagram of the AI process chain INPUT 6 interfaces around 80 records per run RULES Pre-filter hard criteria no model cost AI AI scoring approx. 40 % qualified model per task AI Research facts with sources not assumptions OUTPUT Document package in about 3 minutes every 20 minutes Across everything: quality gate with 140+ tests · targeted self-correction · duplicate detection · failover · monitoring with push alerts
How the AI process chain works. Schematic, as of 09/2026.

Live demo & artefacts

The demo runs on this site.

The assistant in the corner is not an off-the-shelf chat widget but a production n8n workflow that I built and run. Give it a try.

Live demo · n8n workflow in production

What happens to your question

  • my own server checks origin, length and rate of the request
  • it attaches the access key, which the browser never sees
  • the n8n workflow calls a language model with fixed facts
  • the answer comes back with jump links and is labelled as AI

Artefacts and documentation

Part of every project

  • workflow documentation and architecture diagram
  • operations runbook, checklists and UAT
  • handover (transition) to operations and the business
  • versioned, documented deployments via Forgejo and Coolify
  • project metrics, measured continuously rather than estimated

Examples

What you can see here

  • the case studies above with starting point, decision and result
  • the diagram of the AI process chain
  • this assistant as a live example
  • a look inside the workflows, happily on a shared screen in a call

How I work

I run AI projects like any other project: with a metric, clear ownership and a plan for when things fail.

  • 1 · Value before technology

    First the use case and the metric that defines success. The model is chosen afterwards, not before.

  • 2 · Rules as code

    Whatever can be measured is enforced in code. A language model gets instructions, but a program checks the quality bar.

  • 3 · Human in the loop

    Where a mistake gets expensive, approval stays with a person. The path there is automated, the responsibility is not.

  • 4 · Design for operations

    Monitoring, failover, cost and data protection belong in the first draft. That is the part prototypes leave out.

  • 5 · Report honestly

    A measurement that disproves an assumption belongs in the status report too. The silent exception is the costliest risk.

  • 6 · Translate

    Between business, engineering and management there is usually a translation to be made. I make it, for AI just as for hardware.

Foundation

More than 5 years of IT project management. Verifiable, not decorative.

As IT project manager and technology consultant at INVENSITY for automotive and semiconductor projects, before that as co-founder with full P&L responsibility. The complete background is in my CV and on LinkedIn.

Development budgets
0€M

shared responsibility, software and hardware projects end to end

Backlog reduced
0%

through redesigned Jira workflows and fixed routines

Team size
5–8people

development and DevOps, internationally distributed

Collaboration
EU – JP

asynchronous across time zones as the standard

Training

Theory and practice in the same period.

With HERZBLUT Business Training, a certified education provider under the German AZAV accreditation. What the course covers runs in parallel in the projects above.

Completed 09/2026 · 90 of 90 points

AI Agent Development & Business Process Automation with n8n

140 teaching units plus 75 hours of practice.

  • workflows, APIs, authentication and data formats
  • knowledge bases and RAG
  • error handling, logging and monitoring

In progress · completion 10/2026

AI Analyst

160 teaching units. Assessing the potential and limits of AI in a company and choosing tools for good reasons.

In progress · completion 11/2026

AI Manager, advanced course

160 teaching units. Planning, running, monitoring and legally safeguarding AI projects.

Contact

If this fits, I look forward to hearing from you.

The assistant answers questions about projects, methods and tools right away. It only knows what is on this site and in my CV.