AI Engineer & Data Scientist · Berlin, Germany
I build LLM, RAG and document-intelligence pipelines end to end — from an open-ended problem statement to something running in production. Most recently at a VC-backed French Silver Tech startup, where I designed and built three LLM / vision-language document pipelines, all three now in production use.
Before tech I worked in international arts, media and audio post-production across four countries and four working languages. What carried over is the way of working: learn the domain, take an ambiguous problem, and see it through.
AWS Certified Cloud Practitioner. EU citizen, based in Berlin.
AI & LLMs — Prompt engineering · RAG · LLM-as-a-Judge evaluation · LangChain ·
LangGraph · MCP servers & clients · Hugging Face · Vector DBs (Pinecone, Chroma) · LoRA
Machine Learning — Regression · Classification · Clustering · Deep learning
(CNNs, LSTM, Transformers) · NLP · Statistics · A/B testing
Document intelligence & OCR — Vision language models · PyMuPDF · Tesseract ·
image preprocessing
MLOps & deployment — MLflow (incl. MLflow GenAI) · Docker · Kubernetes · CI/CD ·
Flask · Django · Celery · LangSmith · Streamlit · Gradio
Cloud & data engineering — AWS (EC2, S3, RDS, Lambda, SageMaker, EventBridge,
ECR, IAM) · Azure ML · GCP · Airflow · dbt · Snowflake · MongoDB · Redis · Neo4j
Languages & data — Python · SQL · R · Bash · JavaScript
| Project | What it is | Stack |
|---|---|---|
| MEMO — AI Meeting Intelligence Agent | Transcribes recordings with speaker diarization, stores embeddings, answers RAG queries for summaries, insights and action items. Custom MCP tools and servers; Docker-deployed on Hugging Face Spaces. | WhisperX, LangGraph, OpenAI, Pinecone, LangSmith, Docker, MCP |
| EUR/USD Prediction — Production MLOps Pipeline | Automated ML pipeline on a decoupled cloud-native AWS architecture with MLflow tracking, a containerized Flask prediction API and an interactive EDA web UI. Deployment fully scripted via AWS CLI + Bash. | Flask, MLflow, ARIMA, LSTM, scikit-learn, AWS, Docker |
| GDPR & AI Compliance Assistant | Multilingual RAG chatbot for German GDPR compliance, with vector retrieval, real-time monitoring and evaluation scores. Deployed web interface. | LangChain, Pinecone, Streamlit, LangSmith |
| Apple Silicon NN Training Optimizer | Self-initiated toolkit of heuristics for estimating safe batch size, sequence length and memory use when training on memory-constrained Apple Silicon GPUs, plus a constrained search for the best RNN config in a fixed budget. | TensorFlow, Apple Metal, NumPy, psutil, pytest |
| E-Commerce Analytics Pipeline | ELT warehouse on Snowflake: 17 dbt models in a layered build (staging → star schema → marts), 51 dbt tests green on a full build, orchestrated by three Airflow DAGs with retries and a daily schedule. | Snowflake, dbt, Airflow, AWS S3, SQL |
| NLP Fake News Detection | Text classification pipeline with NLP preprocessing and supervised ML. | NLP, scikit-learn |
| Real Estate Price Prediction | Regression model for housing prices, evaluated with cross-validation and error metrics. | pandas, scikit-learn |
| Berlin Rent vs. Inflation | Statistical hypothesis testing on housing costs and inflation, with supporting visualizations. | Python, statistics, matplotlib/seaborn |
| SQL iMDB Movies Database | Schema design with indexing and analytical queries for movie analytics. | SQL |
| Happiness Index Dashboard | Interactive dashboard visualizing global happiness KPIs. | Tableau |
Data Science & AI Diploma — Ironhack, Berlin (2025–2026)
AWS Certified Cloud Practitioner (11/2025)
Generative AI for Data Analysts — IBM (10/2025)
Certified SQL Developer, Professional Level — W3Schools (05/2025)
Before tech
State-Certified Sound Designer — Berlin (systems thinking, DSP, complex data-stream management)
Master's-level degree in Fine Arts (Media Art) — Academy of Fine Arts Leipzig (HGB) (German Diplom)
Spanish (native) · German (C1/C2) · English (C1) · Italian (C1)