Enterprise AI Systems Engineer · Europe

AI systems built for real operational constraints.

Private RAG, document intelligence and compliance infrastructure for organisations that cannot compromise on data, evidence or control.

View selected systems

Private RAG · Document intelligence · ComplianceScroll to inspect systems

What I build

Systems where retrieval quality, source attribution, privacy posture and failure handling matter as much as model quality.

Every system I ship enforces evidence boundaries, validates outputs deterministically, and keeps sensitive data inside the organisation's perimeter. The goal is not another AI demo—it is production infrastructure that teams can audit, explain, and trust.

Selected systems

Selected work

01 — Enterprise validation system

AEC Spec Validator

AEC teams must prove that BIM models satisfy evolving specifications, but requirements, model facts, findings and approvals are usually fragmented across files and manual reviews.

Role

Evidence-first AEC engineering

Key decision

Deterministic IFC / IDS validation

Impact

A production-style validation workflow where every conclusion can be traced from source clause to rule, model evidence, finding and human decision.

Read case study

02 — AI call-intake platform

Anruvo

German service businesses lose requests when calls arrive during field work, after hours or while the team is already occupied.

Role

Vertical voice AI product

Key decision

Voice extraction + deterministic business rules

Impact

A focused SHK product with a functional simulator, tenant-isolated workspace and signed Realtime SIP entrypoint, built around controlled handoff rather than autonomous guesswork.

Read case study

03 — Sovereign RAG platform

FileGPT.dev

Companies want useful answers from internal documents without public files, cross-tenant leakage or sending entire documents into a model context.

Role

Private knowledge systems

Key decision

Private ingestion + scoped pgvector retrieval

Impact

A credible Sovereign AI showcase with authenticated APIs, production rate limits, readiness checks and documented operational boundaries.

Read case study

04 — Security questionnaire workflow

TrustRespond.ai

Vendor security questionnaires slow B2B deals because teams repeatedly map policies and evidence into large spreadsheets while preserving context and formatting.

Role

Reviewable compliance automation

Key decision

Document-to-XLSX RAG pipeline

Impact

A functional private prototype that makes sources, gaps, confidence and approval visible instead of presenting AI-generated answers as final truth.

Read case study

How I build systems

AI workflow architecture

A lightweight view of how I structure production-ready AI systems: deterministic flow control around LLM intelligence, with validation and observability built in.

ACTIVE / 01

User input

Requests enter through typed interfaces with schema-safe parsing and context capture to avoid ambiguity at the edge.

Reliability cues

Traceable steps and outputs

Validation before delivery

Policy-aware orchestration

Engineering evidence

Built systems, shown without theatre.

No simulated latency, invented confidence scores or staged terminal output. This section documents what is deployed, how it is structured and where it can be verified.

Live productPrivate source

ComplianceRadar.dev

A public compliance workflow, not a concept demo.

ComplianceRadar turns a submitted website into structured, evidence-linked findings for EU AI Act, GDPR and ePrivacy review. The product is publicly accessible; the implementation remains private.

Open live product Read technical case study

Directly verifiable

Public scanner and live product
Technical case study with architecture and trade-offs
No synthetic performance or accuracy claims

System path

URL intake

Deterministic checks + scoped model analysis

Evidence-linked findings

Dashboard + public verification

Engineering decisions that can be examined

Evidence before verdicts

Findings preserve the detected page signal and remediation context instead of presenting an unexplained score.

Layered scanning

Fast HTML parsing is backed by a browser fallback for websites whose relevant content is rendered client-side.

Clear trust boundaries

Authenticated dashboard state and public scan verification are separate surfaces with distinct responsibilities.

Operational safeguards

Request protection, scan de-duplication and monitored-domain workflows are part of the product architecture.

Project status

The rest of the portfolio, labelled plainly.

FileGPT.dev

Active private build

B2B engineering agency platform (the 'Compliance Honeypot') for Sovereign AI Assessments and private RAG architectures. The repository is private; the case study documents the intended architecture.

View case study

TrustRespond.ai

Prototype / foundation

Security-questionnaire automation under active development. No speed, accuracy or time-saved claim is presented here as a measured result.

View case study

Selected writing

What I learn while building.

Architecture notes from live products and applied enterprise AI work.

View all articles

Latest

On-Premise RAG for Manufacturing: Architecture, Hardware, and Cost

An auditable FastAPI, vLLM, and pgvector reference architecture covering security, GPU sizing, evaluation, and realistic project costs for technical documents.

Read article

Article / 02

RFQ Automation: How Machine Builders Evaluate 200-Page Specs with On-Premise AI in Minutes

Accelerate technical RFQs and specification documents in B2B sales. How local, GDPR-compliant AI workflows free engineers and protect IP.

Article / 03

Why German Mittelstand Avoids Public-Cloud AI – and How Sovereign AI Saves Document Workflows

How German B2B companies automate manual document processes with local, GDPR-compliant on-premise LLMs—without sending data to the public cloud.

PROFILE / 2026

About

I build enterprise-ready AI systems focused on RAG, agent orchestration, and automation under real production constraints.

My approach is validation-first: deterministic rules and system boundaries where needed, with LLM reasoning layered on top for speed and explainability. The objective is reliable outcomes, not demo output.

I work on systems where retrieval quality, citations, abuse controls, rate limits, privacy posture, and maintainability matter as much as model quality.

Most of my work sits at the intersection of backend architecture, data pipelines, and applied AI product engineering for B2B and enterprise use cases.

I came to tech through a non-traditional path, which shaped how I engineer systems: understand the real workflow first, then ship software that holds up in production.

CONTACT / OPEN CHANNEL

Let's discuss what you're building.

Available for AI systems engineering, RAG architecture consulting, and compliance infrastructure projects. Based in Europe, working with teams globally.

Typically respond within 24 hours · Europe-based · English & German