Projects & Technical Write-Ups

A selection of technical projects, architecture deep dives, and applied research.

RSS Feed

Special Feature • Master's Thesis

Schema-Guided KG Construction and Retrieval for LLM-Based Automotive Test Scenario Generation

Officially registered • TUM Chair of Robotics, AI and Real-Time Systems

Comparative evaluation of retrieval strategies (flat vector retrieval, LightRAG, HippoRAG 2) and automated schema-guided Knowledge Graph construction for generating test scenarios from UN Regulation No. 152, benchmarked against 59 annotated test scenarios.

View Thesis Details & Living Log →

Salsuki: Community Management Platform

2024 – Current • Full-Stack, Solo Engineer

A community platform for a non-profit salsa school in Munich with 60+ active users. Built with Supabase and React 19/TypeScript on Vercel. Handles class scheduling, registrations, credits, waitlists, and a levelled learning resource library.

Read more →

LECture-bot: GenAI Course Assistant

Apr 2025 – Aug 2025 • TUM DevOps Course

Designed the RAG and generative AI subsystem for an intelligent course material assistant. LangChain microservice with document embedding, Weaviate vector database, and integrations with cloud and local LLMs.

Read write-up →  •  GitHub Repository

AI Travel Journal Assistant

Apr 2025 – Aug 2025 • TUM Connected Mobility Lab

Full-stack AI-powered travel journal assistant integrating React, FastAPI, PostgreSQL, and Docker. Led GenAI vision microservice for automated photo analysis and timeline creation.

Read write-up →  •  GitHub Repository

TUM IDP: LLM Tooling Framework

Oct 2024 – May 2025 • TUM Interdisciplinary Project

A framework designed to reduce manual effort in transitioning research-stage software tools to professional production applications using LLM models (Python and Gemini).

Read write-up →  •  GitHub Repository

ADLM Medical Image Segmentation

Apr 2024 – Aug 2024 • TUM Chair for AI in Medicine

Research project investigating resolution differences and state-of-the-art segmentation techniques on 3D medical images using PyTorch, Deep Learning, and nnU-Net.

Read write-up →  •  GitHub Repository