I build production AI systems end to end: retrieval, LLM applications, AI agents and the data pipelines and evaluation underneath them. I care most about the parts that decide whether an AI feature can be trusted in production: honest evaluation, quality bars fixed before results are seen, human review for risky decisions, guardrails, cost tracking per call and reproducible runs.

Since my MSc in Data Science I have shipped independent AI systems that are live and public, from a hybrid RAG system over 8.84M passages to a multi-tenant AI support SaaS built in seven days, and a GraphRAG engine on Neo4j that I benchmarked against a strong hybrid retriever with a pre-registered, run-once test.

I work like a product engineer: understand the problem first, ship the simple version, measure it, then harden what production exposes.

Open to: AI Engineer, LLM Engineer, NLP Engineer, ML Engineer and AI & Data Engineer roles across Germany. Available immediately, open to relocation.

Languages

Frontend

Backend & Databases

AI / ML

Cloud, DevOps & Tooling

Domain	Proficiency	Details
Retrieval-Augmented Generation		Hybrid BM25 + dense retrieval, weighted fusion, cross-encoder reranking, GraphRAG on Neo4j
LLM Evaluation		Gold sets, EM/F1, RAGAS, closed-book baselines, bootstrap confidence intervals, hand checks
AI Agents & Workflows		LangGraph multi-agent pipelines, ReAct agents with tools, n8n workflows, human-in-the-loop review
LLM Engineering		Structured JSON output, prompt versioning, provider pinning, caching, cost tracking, guardrails
Knowledge Graphs		LLM entity and relation extraction, entity resolution, Neo4j
Fine-Tuning		QLoRA on TinyLlama 1.1B, HuggingFace PEFT
MLOps & Deployment		Docker, Kubernetes, AWS EC2, CI/CD, Prometheus

AI Engineer  |  Independent Projects, Berlin
_{April 2026 - Present}

Building and shipping production-style AI systems end to end, with a focus on retrieval, agents, reliable LLM pipelines and honest evaluation.

• Built a GraphRAG engine on Neo4j that raised exact match from 0.57 to 0.79 over a hybrid RAG baseline on 375 held-out test questions (pre-registered, run once, 1.82 USD total LLM cost)
• Built an LLM email intake service with a job queue, retries, guardrails, a human review queue and 175 tests
• Shipped a multi-tenant AI SaaS in 7 days with auth, streaming, CI tests and monitoring
• Released open-source tooling that gives AI agents clean, typed data

MSc Thesis, Retrieval and LLM Systems  |  University of Europe for Applied Sciences, Potsdam
_{March 2024 - March 2026}

MSc thesis: AI-Powered Retrieval-Augmented Assistant for Evidence-Based Question Answering.

• Hybrid retrieval over 8.84M passages reaching 93% Recall@10; alpha 0.70 beat alpha 0.5 by 2.9 pts (p = 0.002) and BM25 only by 11.4 pts
• RAGAS evaluation pipeline, QLoRA fine-tuning and deployment on AWS EC2 and Kubernetes

learning:
  - Graph retrieval and path scoring on Neo4j
  - Model Context Protocol (MCP) and agent tool integration
  - Azure AI and responsible AI governance

building:
  - Applying the GraphRAG benchmark method to agent evaluation
  - A governed AI agent with tool access rules and audit logging

exploring:
  - Evaluation methods for agents and multi-hop retrieval
  - Cost and reliability patterns for production LLM systems

open_to:
  - AI Engineer, LLM Engineer, NLP Engineer, ML Engineer roles
  - Germany, on-site, hybrid or remote, available immediately

Measure honestly, ship the simple version first, and harden what production exposes.