Projects
Things I have worked on.
Most of the projects were created with friends, for example, entel and activity. But each of them is looking forward to see some new contributors 😀
Master's Thesis: AI Agents for Truffle Performance Optimization
A layered knowledge enhancement architecture for AI coding agents to guide GraalVM/Truffle language developers through performance optimization.
Abstract
Building Domain-Specific Languages with the Truffle framework on GraalVM promises competitive performance, but achieving it requires Truffle-specific profiling and compiler knowledge that most language developers lack. Profiling tools like the CPU Sampler, Trace Compilation, and the Ideal Graph Visualizer demand deep framework expertise to interpret, leaving developers without effective guidance during performance optimization. Existing AI coding agents attempt to address this gap but optimize through static reasoning without benchmark validation, often introducing regressions rather than improvements.
Architecture
The thesis proposes a layered knowledge enhancement architecture for AI coding agents with three levels: Tool Assistance via agent skills for individual profiling tools, Process Assistance via sub-agents for optimization process phases, and Full Orchestration. The architecture is implemented as a Claude Code plugin combining agent skills, sub-agents, and an MCP server for the Are We Fast Yet benchmark framework, incorporating domain knowledge sourced from a contextual inquiry interview with a Truffle expert.
Results
On a Degraded Implementation, the Enhanced Agent achieved a significantly higher speedup (U=96, p<0.001, r=0.92). Both user study participants resolved a performance issue with the Enhanced Agent while neither succeeded with the manual CLI setup. These findings establish that AI agents equipped with domain-specific profiling knowledge can lower the barrier for language developers to achieve competitive performance.
fCM.js
Modeling Tool with Design-Time Support for Fragment-Based Case Management.
Goal of the project
It aims at supporting users at design time by providing a joint, visual user interface for all artifacts and by integrating automated guideline checking based on fCM guidelines.
My work
We added a execution interface to the tool, which allows users to execute their models and see the results in real-time.
Future
In the future, we want to export and import Object Centric Event Logs.
antonykamp.de
A minimal nextJS template you're looking at.
Future
I want to polish this website that you can use it as a nextjs-template in the future. This includes adding unit tests and elements of the GitHub GraphQL API to the contributions page.
QuizBot
A Telegrambot to create and attempt to quizzes easily.
What is QuizBot?
QuizBot is a Telegram bot with which you can create and attempt quizzes. A quiz is a collection of possibly random questions.
Deployment
You can try QuizBot on your own! It's currently deployed on Heroku with a MongoDB database saving your and your friends' quizzes.
krisenchat gGmbH
In a crisis? Text us. We'll help you!
Goal of the project
We support young people in crisis by offering chat counselling with professionals. Our team builds the platform and tooling that connects helpseekers with counsellors.
What I work on
Led the replacement of our legacy authentication with BetterAuth, migrating the full user base without downtime. Designed and built a state-machine-driven chatbot waiting room that routes helpseekers to the right counsellor. Contributing to our migration to GCP and our Terraform-based infrastructure-as-code setup.
Future
Continuing to harden the platform and improve the counsellor webapp to make counselling more effective.
AutoDot
Tuning Quantum devices with a flexible statistical approach.
Goal of the project
Quantum devices used to implement spin qubits in semiconductors are challenging to tune and characterize. AutoDot is a statistical approach to tuning such devices flexibly.
Future
We're currently working on a new unit vector sampler using the Covariance Matrix Adaptation Evolution Strategy.
bookkeeper-portal
A web application for keeping track of items and loaning them out.
Goal of the project
Created in the Scalable Software Engineering course at the HPI in Potsdam. Therefore, the main goal was learning working in a big team and scrum.
TicTacToe-Robots
The battle of TicTacToe robots
What you can do
You can create your TicTacToe-Robot and fight against mine or yourself.
DoubleDotDetector
Detection of Doubledots using Viola-Jones Algorithm
Goal of the project
You can generate and prepare data containing doubledots to train the Viola-Jones Algorithm, with which you can detect them in pictures. This project is in cooperation with the University of Oxford.
Future
In the future, we're working on scalable stages in the Viola-Jones Algorithm.
ELEA
Build your own Evolutionary Algorithm.
Goal of the project
You can use ELEA to demonstrate evolutionary behaviors to students to help their understanding. Create your own algorithms quickly using Blockly.
Future
In the future, we're working on plotting the course of different variables to compare algorithms.
entel.me
Share your needs!
Goal of the project
Replace the unnecessary walk to the next supermarket by a new shoppinglist in entel. If you are going to the store, help others close to you by accepting their shopping list and bring them the products they need!
Future
In the future, we want to make the app more intuitive, for example, by replacing mails with push notifications.