Matthias N.

I build AI-powered tools that replace manual document processing, data extraction, and business workflows, so they run automatically with human-level accuracy. Most automations break when they hit messy real-world data: scanned PDFs, inconsistent email formats, multilingual documents, unstructured inputs. I build systems that handle this reliably by combining targeted AI (small, focused LLM calls for judgment tasks) with deterministic code for everything else. The result is automation that works on the first run, not after weeks of debugging. What I build: → Document processing pipelines, ingest PDFs, invoices, contracts, receipts, and emails. Classify, extract structured data, validate, and export to your systems. → AI-powered data extraction — OCR with adaptive preprocessing, field extraction with confidence scoring, cross-field validation. Handles scanned documents, faxes, and mixed-quality inputs. → Workflow automation, connect APIs, databases, and services into end-to-end pipelines. n8n workflows, Python CLI tools, or custom integrations with Microsoft 365, Google Workspace, Slack, and CRMs. → Compliance and reporting tools, automated tax reporting, regulatory monitoring, audit trail generation. Built for German/EU compliance requirements. Tech: Python, JavaScript/Node.js, REST APIs, Claude API, OpenAI API, Microsoft Graph API, PostgreSQL, Tesseract OCR, n8n, OAuth2 Background: Automation engineering in the automotive industry, followed by hands-on AI development work. I understand both industrial process design and modern AI architecture, which means I build tools that are reliable and maintainable, not just clever demos. Every deliverable comes with documentation, tests, and a handoff that ensures you're never dependent on me to maintain it.