AI engineering
LLM applications · Claude · OpenAI · LangChain · structured outputs · retrieval · evaluation · Vertex AI
Applied AI Engineer · Germany
I build AI products that earn their place in real workflows.
My work sits between software engineering, machine learning, and product delivery. I take unclear problems, shape the technical path, and build the data pipelines, backend services, AI workflows, and interfaces needed to put a useful system in people’s hands.
Three years across applied AI, backend engineering, cloud infrastructure, and data-intensive products.
18,507Labeled samples in the crop dataset
60%Less processing time in a Python workflow
15%Improvement in internal model accuracy
Selected technologies
A practical stack, chosen to ship.
Python is the center of my work. Around it, I use managed cloud services, clean APIs, and the smallest useful amount of AI orchestration. The stack changes when the problem does.
Backend
Python · Django · Flask · REST APIs · SQL · event-driven services · validation · testing
Cloud and data
Google Cloud · AWS · BigQuery · Cloud Run · Pub/Sub · Firebase · Docker · GitHub Actions
Product surfaces
Flutter · Android · Next.js · TypeScript · internal tools · field applications · data products
Flagship project · Agricultural intelligence
RaDa Intelligence
A working platform that joins satellite imagery, weather data, and field observations to help agricultural teams see crop conditions earlier and act with better evidence. I designed and built the system across data ingestion, cloud architecture, model workflows, backend services, and field-facing products.
Selected work
Systems, not pitch decks.
The work below covers a field intelligence platform, a transparent civic tool, and an internal automation programme. Different domains. The same discipline: understand the workflow, build the right boundaries, and measure what changed.
RaDa Intelligence
Satellite-to-field intelligence with geospatial pipelines, model training, Android data collection, and web tools for agronomists and cooperatives.
Read the engineering case study →Wahlkompass
A privacy-first voting information tool. It runs entirely in the browser, uses deterministic scoring, and links every position back to its evidence.
Open the live preview ↗Workflow automation
Python automation and a Django application that replaced manual reporting work, cut processing time by roughly 60%, and made the tooling available across offices.
See the experience summary →How I work
Useful beats impressive.
I like ambitious systems. I’m equally interested in the unglamorous parts that keep them alive: clear ownership, sensible fallbacks, useful telemetry, and an honest answer when the evidence isn’t strong enough.
Find the point of friction.
I map the users, decisions, data, and failure cost before choosing a model or framework.
Small services, explicit contracts.
Data ingestion, inference, business rules, and user surfaces should evolve without dragging one another into a rewrite.
Prompts aren’t a QA plan.
I use validation, representative cases, human review, and operational metrics to learn where the complete workflow fails.
Real use changes the design.
A small release with real feedback teaches more than a polished demo built around invented assumptions.
For hiring teams
Need an engineer who can own the messy middle?
I’m interested in Applied AI Engineer, AI Product Engineer, Forward Deployed Engineer, and AI Solutions Engineer roles in Germany or remote.