Agentic AI applied where correctness is non-negotiable.

Selected work
Compliance · Document AI

GDPR Risk Analysis System

An AI system that analyzes contracts and processing agreements for data-protection risks — deployable fully on-premise, so sensitive documents never leave the company.

Details

Challenge

Reviewing contracts and processing agreements for GDPR risk is slow, expensive expert work — and most companies do it too rarely.

Solution

An AI system that analyzes documents, surfaces data-protection risks with references to the relevant articles, and guides remediation workflows — deployable fully on-premise so the sensitive documents never leave the company.

Why it matters

Exactly the class of AI German companies want: sensitive data, clear ROI, and architecture that satisfies the Datenschutzbeauftragter.

RAGSelf-hosted LLMvLLMOn-Premise
In Development · Voice AI

AI Phone Agent

A voice agent that answers calls, understands requests in German, books appointments, and hands over to a human when it should — currently in development.

Details

Challenge

Businesses miss calls and lose customers; call centers are expensive and hard to staff.

Solution

A voice agent that answers calls, understands requests in German, books appointments, and hands over to a human when it should — currently in development.

Speech-to-TextLLMText-to-SpeechTelephony API
Live Product · SaaS

PromptPageAI — AI Website Generation

A production SaaS that generates complete, hosted business websites from a prompt — live with paying users, built and operated end-to-end.

Details

Challenge

Small businesses need professional websites but can't afford agencies or the time to build them.

Solution

A production SaaS that generates complete, hosted business websites from a prompt — domain, SSL, SEO basics, and GDPR-compliant hosting included, delivered in 24 hours.

Why it matters

Proof of end-to-end shipping ability: LLM pipeline, payments, hosting infrastructure, and operations, run as a real product with paying users.

LLM PipelineNext.jsFastAPIStripe
Visit the live product →
Security · Multi-Agent System

KQL Generator for Microsoft Sentinel

A multi-agent system that generates, validates, and refines detection queries for Microsoft Sentinel from natural-language intent.

Details

Challenge

Security analysts spend significant time hand-writing and debugging KQL queries for threat hunting and detection engineering in Microsoft Sentinel.

Solution

A multi-agent system that generates, validates, and refines KQL from natural-language intent — one agent drafts, one checks schema correctness against the workspace, one refines based on result quality.

Why it matters

Agentic AI applied to a domain where correctness is non-negotiable — with validation loops instead of blind generation.

LangGraphMicrosoft SentinelKQLAzure
Research · Deep Learning / HPC

3D Medical Imaging AI — Colon Cancer Classification

Custom deep learning models for 3D biomedical image classification, trained on high-performance computing infrastructure.

Details

Challenge

Classifying 3D biomedical image data requires custom architectures and serious compute — off-the-shelf models don't transfer.

Solution

Custom deep learning models for volumetric data with fully automated GPU/HPC training pipelines — from data preprocessing to reproducible experiment tracking.

Why it matters

Demonstrates the deep end of the stack: not just calling LLM APIs, but designing and training neural networks on high-performance computing infrastructure.

PyTorchCUDAHPC/Slurm3D CNNs
Your Project

The next case study could be yours

Start with the AI Readiness Check — and know within two weeks whether your use case is worth building.

Start the Readiness Check →