Built a ~55,000-line Python application platform used by 15?20 traders to prepare and submit day-ahead and intraday electricity bids. The platform hosts four NiceGUI and two Plotly Dash applications plus an admin panel, covers thermal generation, hydro storage and renewables, and integrates nine internal and external systems. Users retrieve schedules, plant parameters, availability and price forecasts, adjust inputs and generate exchange bids in one workflow. The applications replaced Excel-based tooling and users report faster workflows, better usability and broader functionality. 

• Designed and built the platform from scratch using hexagonal architecture and tactical domain-driven design patterns (entities, aggregates, factories, repositories, services and domain events); implemented the first two NiceGUI applications end-to-end and established reusable scaffolding for subsequent applications. 
• Decoupled external integrations through repositories and thin Python SDKs; several connected data sources and services were replaced during the engagement with limited impact on application logic. 
• Strengthened engineering quality with structured and context-aware logging, pre-commit, Ruff formatting / linting and beartype runtime type checking, reducing formatting noise and surfacing lint and type issues earlier. 
• Stabilized and parallelized the pytest suite with pytest-xdist, reducing runtime from ~12 to ~3 minutes (75%) and removing assumptions about sequential test execution. 
• Built production-to-development data synchronization so corner-case failures in numerical algorithms can be reproduced locally from the exact production inputs instead of being reconstructed manually. 
• Established an agent-accelerated development workflow: typically 3?5 coding agents work concurrently in isolated Git worktrees, each able to run and browser-test its own application instance. Used the workflow for test-coverage expansion, dead-code removal, type cleanup, elimination of ~600 deprecation warnings and migration to a substantially broader lint ruleset. 

Delivered data engineering and analytics for a cashback credit-card startup whose partner bank handled banking and transactions. Combined daily bank CSV deliveries with customer-app data to support analysis of onboarding, business performance and customer spending. 

• Took over maintenance and further development of the existing ingestion pipeline as its sole developer. Fixed and extended Apache Beam transformations and ingestion into BigQuery, with scheduled runs triggered through Google Cloud Functions. 
• Resolved Python module-import problems and consolidated duplicated pipeline code into reusable modules. Introduced automated tests and type checking to a codebase that previously had neither. 
• Analyzed onboarding drop-off to identify where customers abandoned the process; built business-controlling dashboards and analyzed performance KPIs and spending by category. 
• Independently translated leadership questions into SQL queries and quantitative analyses, providing previously unavailable evidence for business decisions.

Built a forecasting and capacity-planning solution from scratch for a manufacturer of kitchen-furniture surfaces. Produced monthly production-quantity and machine-hour forecasts through the end of the following year, supporting plant planners in machine and personnel scheduling. 

• Combined sales forecasts entered in a web app with historical data from a Microsoft SQL Server warehouse. Filled gaps in sales input with time-series forecasts by product category and major customer, with smaller customers grouped together. 
• Trained and compared forecasting models with hyperparameter optimization and best-model selection, using seasonal averaging after detrending, Prophet, linear regression, ARIMA, XGBoost and Holt-Winters. 
• Translated projected sales into production quantities using historical product mixes, then derived required operating hours by machine type and month from multi-step production-run data. 
• Iteratively investigated data-quality issues and clarified production records with stakeholders. Built a custom web app visualizing production steps and machine usage as graphs to make product-specific manufacturing routes easier to understand. 

Developed reproducible Python analyses to assess demand for clothing concepts before development and production. The retailer tested concepts in its online shop using digitally rendered product images, informing customers who added an item to their bag that it was not yet available and might be produced depending on interest. 

• Compared customer clicks on concept-product pages with existing products to predict sales potential and inform profitability assessments and production decisions. Quantified prediction uncertainty based on the number of observations. 
• Analyzed first-purchase products and subsequent repeat purchasing to identify which products attracted new customers who returned for further orders. 

Extended an application for exploring scraped web news about business partners, supporting review of potential reputational and compliance concerns and alignment with company values. 

• Refactored the app for extensibility, introducing path-based routing, feature separation and direct links to individual views. 
• Designed a SQLAlchemy database schema for ingested news data. Introduced SQLite for faster local tests and Playwright for end-to-end testing. 

• In order to identify geographical claims hotspots, claims data was cleaned, geocoded and geographically clustered

Established the foundations for machine learning at a medical-equipment manufacturer with a mature business-intelligence team and data warehouse, but no existing machine-learning infrastructure or capability. The engagement combined platform selection, pilot implementations and guidance for building an internal team to assess the business value of machine learning. 

• Defined platform requirements from practical machine-learning delivery experience and compared Anaconda and Azure offerings. Implemented the initial use cases on the selected Azure platform using Azure Machine Learning, GitHub and GitHub Actions, including CI/CD for packaging and deploying data-science web apps. 
• Built pilot use cases covering sales forecasting, delivery lead-time analysis, fuzzy prospect deduplication, repair-volume forecasting and repair-type prediction. 
• Authored a data-science handbook explaining how to develop projects on the platform, alongside criteria for evaluating and prioritizing worthwhile use cases. 
• Helped define roles and hiring profiles for an internal data-science team and supported recruitment. 

• Creation of foundational report based on Google Sheets incl. daily automatic refresh (new customer counts, order counts, average order value, customer retention and order frequency)
• Setup of reporting server (base setup, security, CRON job, documentation)

• Consulted on viability, actionability, data requirements, prediction target selection, feature engineering and machine learning process for two use cases
• Mentored junior data scientists throughout the use case implementation including approach discussion, software selection, best practices, code review

• Validated data extract and drove iterative process for fixing data inconsistencies
  
• Ideated and defined data analysis tasks and goals together with the business and implemented the data analysis
• Built report from analysis results and presented to the business

• Gathered requirements of chemists and specifics of chemical domain (adhesives)
  
• Researched Gaussian processes, Bayesian Optimization and open source implementations thereof
• Built web-app for visualizing experiment data and surrogate function as well as proposing new formulations to test next

• Consulted on potential usage of machine learning and custom UI design to improve controllers ?day-to- day work
• Specified data requirements with BI department and ordered daily data provisioning
• Reimplemented formulas for calculating devaluation figures of slow-moving inventory
• Implemented data preparation and parallelized mass-training of forecast models (multiple 100k models trained)
• Built custom dashboard for quick assessment by controllers, regarding the need for manual adjustment of devaluation figures
• Packaged regular model training and dashboard hosting in Docker and Docker-Compose and deployed dashboard

• Consulted on selection and definition of important KPIs to control marketing activities and get insights into consumer behavior
• Implemented the data pipeline for computing KPIs from consumer database and device log data
• Built Power BI dashboard to visualize data

• Selected relevant data and specified the data extract together with BI team
  
• Ideated and implemented custom features for machine learning and executed standard data preparation (text vectorization, dimensionality reduction, etc.)
• Trained machine learning model and evaluated model performance
• Built and deployed pipeline for daily batch prediction

• Researched known scientific methods in epigenetics
  
• Ideated and tested different test statistics in close collaboration with R&D department to find gene locations responsible for test subjects? physical traits
• Enabled employees of the R&D department by familiarizing them with the Python data ecosystem

• Researched and tested different providers of weather data
  
• Automated weather data retrieval and preprocessing
• Implemented web-based frontend with integrated Google Maps widget and backend with REST API

• Researched binary format specification of obscure AFP MO:DCA file format for print documents (similar to PDF) and extended Python AFP library with features required for project
• Implemented memory-efficient AFP parser to extract relevant data from print documents without requiring loading complete (large) files into memory by utilizing Python generators
• Implemented multiple methods for determining document similarity to identify inefficiencies in company?s print jobs

• Researched relevant technologies, determined their project fit and distributed the required information to the different project contributors
• Designed data format for streaming and consulted developers on integration into web-backend
• Managed tasks, contributors and stakeholders of project towards a successful outcome
• Implemented small Spark ETL job based on the ingested data

• Implemented Spark ETL job with Scala based on KPI report specification
  

• Ideated approaches for model training: feature engineering and prediction target
  
• Trained machine learning model and evaluated model performance