I don't throw AI at problems to see what sticks. I find the use cases where AI - or a plain automation - actually pays off, and engineer them into production. Currently shipping Multi-Agent RAG with MCP at a Big 4 company.
Double-promoted from Associate - skipping a level in twelve months. Building Multi-Agent RAG systems with MCP (Model Context Protocol) for context-aware enterprise automation - and leading firm-wide Trusted AI enablement across finance, automotive and public sector.
Most companies throw AI at the wall to see what sticks. I don't. Four steps from a real business problem to measurable, adopted value.
Start from the business process and score every candidate on ROI × feasibility. If a plain automation solves it, the automation wins - AI is the tool, leverage is the goal.
Tools bound at runtime via MCP with least privilege, answers grounded in RAG, plus a validator and audit trail mapped to EU AI Act controls.
Measure the baseline before shipping, instrument every run - resolution rate, cost, latency - and report the delta in business language with the process owners.
Train champions inside each team, run hands-on workshops on real cases, and loop usage data back into the backlog so adoption compounds instead of fading.
Nine years of shipping - from weekend freelance HTML at 14, through a first-class AI degree, two company foundations, and now engineering trusted AI for the Big 4.
Started shipping websites at 14 for local SMEs - restaurants, craftsmen, small e-commerce. Learned the craft end-to-end: turning a client conversation into a live product. Later joined consult.IN e.V. (Oct 2020 – Feb 2022), a student consulting initiative in Ingolstadt, where client-facing delivery became second nature.
Focused on Computer Vision, Transformers and Reinforcement Learning - with a strong foundation in classical ML and math. Bachelor thesis combined YOLOv8 real-time object detection with XGBoost to predict American football plays from broadcast video - published open-access with grade 1.0 on opus4.kobv.de.
Co-founded Stratek while studying - grew from a two-person GbR to a full GmbH. Shipped 10+ web & mobile products and engineered a computer-vision drone pipeline for aerial inspection.
Joined as Associate. Twelve months later, double-promoted (two levels in one jump) into two parallel roles.
Full-stack across the modern AI delivery pipeline - research, agents, cloud, governance, and the consulting craft on top.
Weekly experiments on MCP, agents, and whatever I'm curious about this week. Live pulled from github.com/linus131313.
loading description text here while we fetch data
loading description text here while we fetch data
loading description text here while we fetch data
loading description text here while we fetch data
I believe AI in the enterprise has to be trusted, explainable, and actually deployed - not just demoed.
A Synthesis of Published Evidence.
A review of six engineering patterns commonly invoked to make language-model agents reliable - structured tool schemas, validator agents, reflection-and-retry, least-privilege tool scoping, failure propagation, and deterministic output formatting. Each pattern is mapped to the strongest controlled evidence in the literature and graded on a strong / moderate / weak / absent rubric.
Bachelor thesis combining real-time object detection with gradient-boosted decision trees to predict American football plays from broadcast video. Grade 1.0, published open-access.
opus4.kobv.deEmpirical Evidence on Dynamic Tool Binding in Enterprise AI.
A quantitative study of the gap between Model Context Protocol adoption velocity and the response cadence of ISO/IEC 42001, NIST AI RMF, and the EU AI Act. Introduces the Tool Graph Capability Score (TCS) - a reproducible metric that places a runtime agent on the same axis as traditional approved applications.
What I do when I'm not shipping code.
Six years of competitive team sport - leading the locker room, calling the defense, coaching the next generation. The best management training you can get in your twenties.
Built "StudyMatch" - a Flutter app for matching students by study goals - end-to-end in 4 weeks. Design, development, App Store + Play Store deployment. Won 1st Place and the Sustainability Award at the BBW Gründerpreis.
Drop a line. I read every message.