React Dependency Injection is a TypeScript package that introduces a robust, type-based dependency injection system for React applications. This solution enables developers to easily manage component dependencies, streamline code organization, and enhance maintainability while harnessing the full potential of TypeScript's strong typing capabilities.

• Call arbitrary Python functions from NodeJS
• Use machine learning models in NodeJS
• Fine-Tune machine learning models from data coming from NodeJS (like Typescript ORMs)
• Text-Embedding from and to your database managed by NodeJS/TypeScript

Semantic Analysis Manager is an app designed to help users manage their natural language processing (NLP) datasets, specifically for semantic analysis tasks. The app enables users to effectively organize their datasets and perform evaluations using state-of-the-art fine-tuned transformers and topic modeling techniques. Through the evaluation process, the app provides in-depth insights into the various features detected per text, as well as statistical data visualizations, including correlations and histograms.

This project demonstrates a practical application of advanced NLP techniques to facilitate dataset management and performance evaluation in semantic analysis tasks.

Auto-CMS is an AI-driven content management system that leverages natural language processing to simplify and enhance user interaction with content management tasks. By allowing users to input their requirements in everyday language, Auto-CMS dynamically generates interfaces for managing various types of content, such as lists, add/edit forms, and more.

In addition, Auto-CMS can automatically construct queries and visually display results in diagrams and charts when users request specific statistics or information. This project showcases the potential for AI to streamline content management processes and improve user experience.

TypeRunner is the by far fastest and most efficient TypeScript compiler written in C++. It compiles TypeScript down to bytecode and executes it with a custom runtime to type-check extremely fast. It's the first TypeScript compiler that compiles to bytecode and does not interpret the code.

With its high-performance architecture its up to 10,000 times faster than the original TypeScript compiler tsc. As a proof of concept its purpose is to show that TypeScript can be type-checked much faster than it is today.

A TypeScript implementation of a cryptocurrency inspired by the Bitcoin blockchain. It uses the same blockchain binary format and algorithm including the transaction scripting language (although a light version of it). Since its binary compatible its possible to create transactions for bitcoin miners and read bitcoin blocks and transactions, as well as to validate transaction input locks + output unlock scripts (aka scriptPubKey/scriptSig).

It's a project to understand the Bitcoin blockchain algorithm in-depth, its binary structure, transaction semantics, scaling limitations, and performance bottlenecks. The miner RPC protocol of this project is different to Bitcoins RPC protocol.

❮

❯

Deepkit Framework is a new high-performance and real-time TypeScript framework for sophisticated isomorphic TypeScript applications. It consists of many high-performance TypeScript libraries that can be used alone or in combination with the framework.

Deepkit is the first full-fledged framework that brings types to the runtime and allows with it completely new ways of writing data driven applications. With its fastest ORM, serialization, and validation implementations its the perfect choice for high-performance applications. It's micro-service oriented architecture allows to scale your application out of the box.

• Runtime type system, serializer, and validator
• Data-Mapper ORM for MySQL, SQLite, PostgreSQL, and MongoDB
• HTTP kernel/router
• RPC framework
• Broker
• And much more

All libraries are carefully tuned, well tested, and written with SOLID principles in mind.

Following screenshots display the ORM browser feature:

❮

❯

Deepkit is a collaborative and analytical training suite for insightful, fast, and reproducible modern machine learning. All in one cross-platform desktop and web app for individuals, corporate or open-source teams. Deepkit ML has one of the most powerful machine learning model debugger with real-time analytics and per-layer visualisations.

A C++ implementation of the famouse Pluribus AI bot - the super human texas-holdem Poker AI bot. The challenge was not only to understand and apply more classical artificial intelligence algorithms like Monte Carlo Tree Search, but also understand and implement counterfactual regret minimization, which was used in Pluribus.

The Poker game engine itself was highly optimized in C++ to allow very fast simulation.

❮

❯

Attestation service for customers for SIGNAL IDUNA. A responsive web application written in TypeScript based on the component library.

❮

❯

Living styleguide based on Sketch and Angular components, based on NPM packages so multiple developers and many teams can work always on newest user interface widgets from the design department.

❮

❯

The challenge was to build a 2D composition engine that allows the user to arbitrary create ad images based on pre-defined constraints. The composition engine is based on WebGL using PixiJS and is as powerful as a regular graphical image editor like Sketch/Figma. The GUI controls around is implemented using Angular.

A layout can be saved as a template with certain constraint and placeholders. A constraint can be for example that a logo can only be placed at the top right and text needs to be placed at the bottom. This makes it possible to write correct templates regarding corporate design and let other people fill the content for marketing campaigns.

Via an HTTP API, other services can create images based on templates by filling out necessary placeholders.

❮

❯

An editor for animated HTML5 ads. It allows a user to write HTML with CSS animations and directly preview the animation in the browser. A feature called placeholder allows the user to mark certain as editable by other users converting the HTML ad into a template.

Other users can then open a template and fill out those placeholders why not modifying any HTML/CSS code. An extraction features allows to compress and export the HTML ready to be published to ad campaign tools.

A JavaScript package for Node's npm, that allows you to work locally on multiple packages that depend on each other.

Replacement for npm link done right for local development of multiple packages that are based on each other. It symlink root files using a watcher and excludes "peerDependencies", instead of using symlinks on the whole root. This makes it possible to really work locally on multiple packages that depend on each and have devDependencies checked out.

Pesto is a high-performance GUI framework in C++ highly inspired by CSS and HTML, using Skia as rendering engine. Including a dev-tools like interface to debug the DOM and styling structure.

It's a project to understand how browser's UI engines work in detail and what performance implications there are. It's not finished and probably never will, since the use-case from back then is not existent anymore.

AETROS was building an application called AETROS Trainer, an end-to-end deep learning application and neural network training suite, that helps developer to use machine learning, especially deep learning (neural networks) to solve problems using artificial intelligence without the need of having knowledge of the deeper mathematical principles behind neural networks.

Additional it was a general purpose monitoring solution for analysing the training process of modern machine learning pipelines.

❮

❯

A PHP library with several implementations of a Topological Sort (topSort). In additional to the plain sorting algorithm it provides several implementations of a Grouped Topological Sort, means you can pass items with a type which will be grouped together in the sorting. With its implementation of using strings instead of arrays its over 20x faster than regular implementations.

An acoustic amp for iPhone 5 and docking station in a futuristic style yet functional design.

The iPhone can be charged via the user's charging cable, which is clamped into the device in the holes provided.

The design was created in Autodesk Inventor, the prototype tested using 3d printing on a Renkforce RF1000, and injected moulded in China.

A salary comparison website for the company Xing, a Hamburg-based career-oriented social networking site.

I developed the websites using Symfony PHP.

Element Queries is a polyfill adding support for element based media-queries to all new browsers (incl. IE7+). It allows not only to define media-queries based on window-size but also adds 'media-queries' functionality depending on element (any selector supported) size while not causing performance lags due to event based implementation.

PHP-PM is a process manager, supercharger and load balancer for PHP applications. It's based on ReactPHP and works best with applications that use request-response frameworks like Symfony's HTTPKernel. The approach of this is to kill the expensive bootstrap of PHP (declaring symbols, loading/parsing files) and the bootstrap of feature-rich frameworks. See Performance section for a quick hint. PHP-PM basically spawns several PHP instances as worker bootstraping your application (eg. the whole Symfony Kernel) and hold it in the memory to be prepared for every incoming request: This is why PHP-PM makes your application so fast.

Back then it was a game-change to execute PHP in a long-running process with only a few changes and get in return extraordinary high-performance.

Simple reaction-based game for iOS. The user needs to unlock a lock using his reaction skills. He can play in a unlimited fashion and crack the global high score, or play against other players on the same device.

An enterprise and high-speed open-source Content-Management-System (CMS) / Application Framework based on the Symfony framework with a full RESTful API, a user friendly administration interface and rapid application development (RAD) framework using Composer, Propel and other rock-solid libraries.

With a very sophisticated extension system, GUI for inline editing, and file manager, it was back then one of the most advanced enterprise CMS on the market.

❮

❯

A high-performance real-time sport time-tracking software for marathons and other races. The server is written in pure C and the administration in PHP with a HTML5 interface and a RESTful JSON API. The hardware that actually tracks a runner is ChronoTrack, a RFID tracker.

The server itself is extremely fast and calculates several hundred thousand times/second and provides the result direct live through a HTML5 WebSocket interface to mobile devices (iPad web client). Was is use in several german marathons and in Letlant Marathon, one of the world biggest Marathon with over 16.000 participants.

A highly customizable and blazing fast ORM library for PHP 5.5+.

I primarily implemented PostgreSQL and SQLite migration support, fixed tons of bugs, increased performance, implemented the website and sandbox, improved the automated unit test framework, and started with the implementation of a new Propel3 based on a data-mapper approach.

A enterprise open-source Content-Management-System and Content-Management-Framework with a full RESTful API written in PHP and Mootools.

The graphical interface was miles ahead of its time. It consisted of a inline content editor and feature-rich file manager as well as a rapid application framework to allow CRUD interfaces within minutes.

Mindaffect was a german full-service marketing and web development agency with eight employees and several freelancers. We did marketing campaigns, corporate websites, custom application development, and custom interface for third party applications.

This position came along with close contact to customer and solution- orientated thinking, working with big companies like Hubert Burda, Tesa, Tomorrow Focus and Mercedes Benz.

Social website for party-people with (back then in 2008) over 300.000 registered people.