Back
Projects
GitHub Projects
- JavaScript pi-graphite
Graphite stacked branches as one tool for Pi and Oh My Pi agents.
- TypeScript pi-diagram
Makes Pi and Oh My Pi agents actively explain complex topics with token-efficient D2 diagrams.
- JavaScript pi-herdr-worktree
Herdr worktree workspaces as one tool for Pi and Oh My Pi agents.
- Rust herdr-workspacer
Find projects with zoxide, then switch or create Herdr workspaces.
- Lua herdr-context.nvim
Send Neovim buffers and visual selections to coding agents in Herdr.
- Rust beez
Run coding agents and shell commands as sandboxed tasks with YAML workflows, logs, and schedules.
Research Projects
- Generation of Naturalistic Traffic Rule Violations Using Imitation Learning
Master's Thesis at TU Munich — Using Generative Adversarial Imitation Learning to generate naturalistic traffic rule violations for autonomous driving testing.
- Assessing the Similarity of Traffic Scenarios Using Graph Neural Networks
Research project at TU Munich — Using Deep Divergence Graph Kernels (DDGK) to learn graph-based similarity embeddings for traffic scenarios, enabling clustering and classification.
- Motion Planning with Temporal Logic Specifications
Seminar work at TU Munich — Reviewing integrated task and motion planning approaches that use temporal logics (LTL, CTL, STL) to handle complex specifications such as coverage, sequencing, and temporal ordering for robotic systems.
- Motion Planning Feasibility Checker
Developed the feasibility checker module of the CommonRoad Drivability Checker Toolbox for simulation and feasibility evaluation of automated vehicle trajectories.
- Clustering Similar Traffic Scenarios
Research project at TU Munich — Using a modified unsupervised Random Forest algorithm to automatically find categories of traffic scenarios for autonomous vehicle validation.
- Emotion Analysis of Meetings Using Facial Expression Recognition
Bachelor's Thesis at Ege University — Developed a real-time facial expression recognition system to provide speakers with instant feedback on listener interest levels.