I'm an AI & Machine Learning Engineer with a background in backend software engineering, specializing in LLM systems, autonomous agents, RAG pipelines, and scalable Python applications. Currently pursuing an M.Sc. in Artificial Intelligence Engineering at JKU Linz while turning complex AI ideas into reliable, production-ready software.
The problem with most data analysis tools is that you need to know what questions to ask. DataPilot takes a different approach, given a CSV file, the agent plans its own analysis steps, writes Python code to execute them, retries on failure, and produces a structured PDF report with charts. The interesting engineering challenge was building a reliable code execution loop with self-correction, so the agent recovers from runtime errors without human intervention.
Built a RAG application that lets users query their own documents and receive answers grounded in the source material, with citations pointing back to specific passages. The pipeline covers document ingestion, chunking, embedding, vector storage, retrieval, and response generation. Packaged with a FastAPI backend and Streamlit frontend, deployed via Docker. The main challenge was keeping responses grounded in the uploaded documents while maintaining low retrieval latency. The pipeline combines document chunking, vector search, and citation generation so users can trace every answer back to its source.
Trained and optimized a YOLOv8 model for real-time license plate detection, with the goal of making it fast enough for live camera input. The focus was on the inference pipeline, getting latency down to ~3.5ms on NVIDIA T4 while keeping accuracy high. Evaluated using mAP50 and F1-score, achieving 98.4% and 96.6% respectively.
Built a MARL environment where multiple agents must coordinate to solve warehouse tasks without being able to observe the full state. The core question was how agents develop cooperation strategies without explicit communication. Studied the effect of different reward shaping approaches on coordination quality and training stability across several algorithms, and analyzed where independent learning breaks down at scale.
I enjoy building complete systems that solve real problems.
For me, a successful AI application is more than an accurate model. It also needs clean architecture, reliable APIs, maintainable code, reproducible workflows, and a user experience that makes the technology genuinely useful.
Whether I'm building an autonomous agent, a RAG system, or a computer vision application, I approach every project as a software engineering problem first. AI is one component of the solution, not the entire solution.
I value simplicity, modular design, measurable performance, and building systems that someone else could understand, maintain, and extend.
A multimodal RAG platform that can understand technical documents, financial reports, tables, charts, and images using open-source LLMs and vision-language models.
Expanding the autonomous data analysis agent with improved planning, richer visualizations, and support for larger analytical workflows across multiple datasets.
Exploring cooperative reinforcement learning strategies for warehouse optimization under partial observability, focusing on coordination at larger agent populations.
I'm looking for roles where I can build AI systems that actually get used, I'm looking for roles where I can design and build complete AI-powered software systems, from backend services to intelligent applications. If you're working on something in AI engineering, LLM systems, backend development, or applied ML, I'd like to hear about it.