Roland Polczer

AI Engineer · Unstructured data and in-house AI tools

Modern AI for unstructured data.

Much of a company's most valuable data already exists — scattered across Excel files, PDFs, scans, email threads and internal documents — but not in a form that people or software can use reliably. I build systems that turn it into verified, traceable data. I also provide independent advice on where AI can deliver real value and develop focused in-house tools for research, knowledge work, translation and automation. My work draws on 25 years of experience with production data platforms and several years building autonomous AI systems that use real tools and real data.

See the work ↗ Get in touch ↗
From unstructured sources to verified dataIndependent AI adviceIn-house tools built for the long term
Explore

What I do

01

Less manual work through reliable automation

Quotes, delivery notes, invoices, forms and follow-up emails — the information already exists, just not in a usable format. I build tailored automation that turns it into verified data: known formats are handled with fixed rules, everything else with AI, and every value is checked against a defined schema. People stay in control where judgement matters; the system takes care of repetitive data entry. It is inexpensive to run, straightforward to maintain and eliminates hours of manual data entry.

Less manual data entryHuman judgement where it mattersSimple to operatePredictable costs
Get in touch ↗
02

Where AI can actually pay off

Before anything is built, it is worth establishing whether it should be built at all. I start with your processes, not the technology: which tasks consume the most time, where errors arise that become costly later, and which steps are structured enough for a system to take over reliably. The result is a short, prioritised list covering effort, expected value and prerequisites — not an 80-page strategy document, but a practical basis for making decisions. I sell no licences and receive no vendor commissions. If you choose to proceed, I can build the solution myself.

Use-case assessmentFeasibilityRoadmapVendor-independent
Get in touch ↗
03

Research agents that keep you informed

Competitors, prices, tenders, regulation and funding — the things that matter to your business keep changing, and no one has time to monitor them all. I set up agents that track the sources most relevant to you, separate signal from noise and deliver a short, fully sourced summary on a regular schedule — without filling your inbox. They cost very little to run in the background, and you decide which topics they follow.

Competitor monitoringMarket and regulationSource-backed summariesYou set the topics
Get in touch ↗
04

In-house AI tools for your team

Purpose-built tools for the work your team actually does: document translation that preserves the original layout, assistants grounded in your organisation's own knowledge, research agents that produce fully sourced briefings, and real-time translation for multilingual meetings. Each tool has one clearly defined job, is tested against real cases and can run on your own infrastructure or in the EU Data Zone.

Document translationKnowledge assistantsResearch agentsReal-time meeting translationOn-premises or EU-hosted
Get in touch ↗
05

Private and EU-compliant AI

Two ways to keep company data where it belongs: run the models entirely on your own hardware, or use Azure AI Foundry deployments in the EU Data Zone, where Microsoft processes prompts and responses within its EU Data Boundary. Your content is not used to train the models, costs remain predictable, and your data protection officer gets a clear, defensible answer.

On-premises AIEU Data ZoneGDPRYour data is never used for training
Get in touch ↗
06

Practical training and leadership advice

There are already plenty of providers offering general AI courses. I focus on formats where subject-matter expertise makes the difference: one-to-one guidance on real tasks, hands-on training for the team that will operate the system I built, and clear briefings for owners and senior leaders — covering adoption options, risks, governance and concrete next steps. That includes saying 'not yet' when that is the honest answer.

One-to-one guidanceTraining on your own systemBriefings for owners and leadersRisk and governance
Get in touch ↗

Core service

How document processing works

Runs as a container in your environment — on-premises or in the EU Data Zone
01

Intake

Excel · PDF · scan · email

02

Identify and extract

known formats with fixed rules, the rest with AI

03

Schema validation

every value checked against fixed rules

04

Human review

required only for uncertain results

05

Verified data

into your system · with a full audit trail

A building block that fits into your existing processes: every document type follows the same flow, while you define the formats, fields and validation rules.

Selected work

  • 01

    CrewTaxi

    Document Intelligence

    For maritime transport providers, crew-change information arrives in multi-sheet Excel workbooks, PDF crew lists, scans and last-minute emails. CrewTaxi ingests every source, validates each extracted value against a JSON Schema, merges updates into a single auditable record for each person and exports a clean CSV for the booking system. Known layouts are handled deterministically; unfamiliar ones are processed with an LLM. Nothing enters the system without passing validation. The same low-cost automation can take over repetitive data entry wherever documents are still keyed in by hand — invoices, orders or master data — which is often where a small team can save the most time.

    Schema-validated LLM outputField-level provenanceAppend-only audit logIdempotent batch runs
  • 02

    In-house AI for SMEs

    Business AI

    Purpose-built internal tools for SMEs: a document translator that preserves layout across PPTX, DOCX and PDF files; RAG assistants grounded in company documentation; deep-research agents that compile fully sourced briefings; and a real-time spoken-language translator for multilingual meetings. All of them run on Azure AI Foundry deployments in the EU Data Zone, so Microsoft processes prompts and responses within its EU Data Boundary and never uses company content to train the models.

    PPTX · DOCX · PDFRAG assistantsDeep-research agentsLive meeting translationEU Data Zone
  • 03

    OpenClaw Seismograph

    Autonomous Systems

    An agent-powered news intelligence platform that monitors the AI agent ecosystem daily, scores model predictions each month and publishes bilingual intelligence reports — all automated through GitHub Actions and Cloudflare Workers. There is no database and no server to maintain, and running costs are close to zero. Designed for a solo operator, it runs itself.

    GitHub ActionsCloudflare WorkersNear-zero running costsLow-maintenance design
    openclaw.lu ↗

Technology

The technology behind the work.

I use every item here in production, every day.

Agents and orchestration

Model Context Protocol (MCP)Custom Python agentsMulti-agent workflowsEvaluation and guardrailsPersonas · memory · routingHermes AgentDeerFlow

Inference and LLMOps

vLLMllama.cppLiteLLM routingUnsloth fine-tuningLoRA trainingAzure AI FoundryOpenRouterGroqSambaNovaGoogle AI Studio

Data and retrieval

PostgreSQLpgvectorSupabaseOracle · PL/SQLMS SQLFirestoreSearXNGCrawl4AIETL/ELT · Azure Data FactoryStar schema · Data Vault 2.0

Media and infrastructure

ComfyUIDiffusion models for image and videoVoice design · text-to-speechDockerGitHub ActionsCloudflare WorkersTailscale mesh networkingUbuntu · WSL

Both worlds

Portrait of Roland Polczer

Enterprise experience: telecoms billing and rating at Deutsche Telekom and T-Mobile, large-scale rollouts, migrations and reporting, cloud data warehouses, and high-volume Oracle pipelines. I know what happens to production data when formats drift, master data quietly goes wrong and a cutover cannot slip. Modern AI means agents, orchestration, retrieval, evaluation and guardrails — technologies I build and operate every day rather than simply read about. Most people in this field know one side or the other. The most interesting work happens where they meet, and that is where I work.

  • Based in Lienen, Germany
  • Working languages: English, German, Hungarian
  • Oracle · MS SQL · Data Vault · ETL
  • Led teams of up to 12
  • BSc in Computer Engineering · BSc in Finance and Accounting
  • Top IT Worker of the Year — T-Mobile, 2002

Writing

I write about where AI development is actually heading — grounded in practice, without the hype.

All posts on Substack →

Contact

Available for new projects

Get in touch.

Email is the quickest way to reach me, and I read every message myself. If you have documents that need to become usable data, or an AI idea that has to hold up in production, describe it in three sentences. I will tell you honestly whether I can help. Engagements often begin with a small, clearly defined project. That is a deliberate choice.

Reply within 24 hours · Free introductory call · Lienen, North Rhine-Westphalia — on site in Münsterland or remote