Data Science · AI Engineering · Hamburg

From raw data to
deployed AI

I build data and AI applications end to end, from the pipeline to the deployed product. Before moving into data and AI engineering, I spent 6.5 years in SaaS sales. That background shapes how I build. I start from the user and their use case, then work backwards to the implementation.

View Projects GitHub Kaggle Tableau Public

Work

Portfolio Projects

LangGraph · LangChain · RAG · Chroma · FastAPI · Claude API · LangSmith · AWS · CI/CD

Agentic Data Analysis Assistant

Click on the video to see the agent in action:

A production-grade agentic data analyst, live on AWS, with Docker containerisation and GitHub Actions CI/CD. Upload one or more CSVs plus optional context files — data cards and schema diagrams — and ask business questions in plain English. The agent translates your question into SQL, executes it against your dataset, and returns answers grounded in actual query results. Built on LangGraph with a two-tier RAG memory architecture; past analyses are stored in a Chroma vector store and retrieved automatically for future questions.

  • Natural language to SQL — the agent translates questions, executes queries, and returns grounded answers
  • Schema confirmation loop — ambiguities, key relationships, and foreign keys flagged and resolved before any analysis begins
  • Two-tier RAG memory — past analyses stored in Chroma, retrieved via similarity search across sessions
  • Multimodal context input — upload data cards, data dictionaries, and schema diagrams alongside CSVs
  • Deployed on AWS App Runner with Docker containerisation and GitHub Actions CI/CD
  • Automated cost protection — budget threshold triggers an SNS kill switch to pause the service
Python LangGraph LangChain RAG Chroma FastAPI Claude API LangSmith SQLite Docker AWS GitHub Actions
Live App (AWS) GitHub
Tableau · Data Storytelling

AI Job Market — The Future of Work

A Tableau Story with three dashboards built from a single dataset, each reframing the data for a different audience: career changers, HR directors, and policymakers

  • Custom calculated fields to drive interactivity and dynamic filtering
  • Career Changers: find future-proof roles by salary, automation risk, and experience level
  • HR Directors: workforce planning and automation risk by role and industry
  • Policymakers: labour market outlook for Germany 2024–2030, with actionable reskilling recommendations
Tableau Python Kaggle
Tableau Public GitHub Kaggle
Python · EDA · BigQuery

Google Merchandise Store — Marketing Channel Analysis

Python/Jupyter Notebook EDA structured as a CMO-level briefing

  • Queried real GA360 data directly from BigQuery using a two-table extraction architecture separating session and hit grain
  • Classified and compared marketing channel performance across conversion and revenue metrics
  • Segmented traffic by region and analysed conversion behaviour across geographies
  • Tracked revenue per session over time to identify seasonal acquisition windows
  • Applied permutation testing to assess statistical significance of a low-volume channel anomaly
Python pandas matplotlib seaborn BigQuery
GitHub Kaggle
SQL · Relational Analysis

Olist E-Commerce — Customer Satisfaction Drivers

SQL analysis of a real e-commerce relational database, structured as an operations team briefing

  • SQL analysis of a real star-schema relational database across 8 tables and ~100,000 orders
  • Data quality phase establishing null rates, join integrity, and delivery data completeness before any analysis
  • Tested four logistics metrics against review scores to identify the strongest predictor of customer satisfaction
  • Attributed delivery underperformance between sellers (dispatch speed) and carriers (transit time)
  • Decomposed geographic performance patterns across state, city, and zip code prefix granularity to identify the correct intervention level
SQL SQLite VS Code
GitHub
Data Science Bootcamp

Rakuten — MLOps Production Pipeline

  • End-to-end MLOps pipeline with intelligent retraining triggers and drift monitoring
  • Containerized microservice architecture deployed on Google Cloud Platform
  • Real-time API deployment with monitoring and alerting via Prometheus and Grafana
Python MLflow Airflow Docker GCP Prometheus
Streamlit App GitHub
Data Science Bootcamp

Rakuten — Multimodal Classification Ensemble

  • Multimodal ensemble combining classical ML, BERT, and VGG-16 for e-commerce product categorization
  • Comprehensive feature engineering across text and image modalities
  • Applied computer vision and NLP techniques to a real-world categorization challenge
Python PyTorch BERT VGG-16 Scikit-learn
Streamlit App GitHub

Capabilities

Skills & Tools

Focus Stack

Python SQL LangGraph FastAPI RAG Chroma Docker AWS GitHub Actions Tableau

Further Tools

Data Science & ML

  • pandas
  • NumPy
  • scikit-learn
  • PyTorch
  • TensorFlow
  • matplotlib
  • seaborn

Data & Cloud

  • BigQuery
  • PostgreSQL
  • SQLite
  • GCP

MLOps & Monitoring

  • MLflow
  • Airflow
  • Prometheus
  • Grafana