My Journey

A timeline of my professional experience and academic achievements.

Work Experience

October 2025 - Present

Master Thesis

Part Time

AIT Austrian Institute of Technology, Vienna, Austria

Built and shipped production AI infrastructure in the AI-Taskforce: an enterprise RAG chatbot over ~2,000 internal QM documents, validated by domain experts and selected for enterprise-wide rollout. Independently evaluated across 2,000 real-world queries — 90% rated Good or better (4.61/5 overall), with top document categories hitting 4.94–4.96/5 and near-perfect source citation accuracy (96–100%). Consistent performance across German and English, and from simple to complex questions.

AIT Austrian Institute of Technology is Austria's largest non-university research institution. Within the AI-Taskforce, I designed and built three interconnected production systems that form the applied infrastructure for my master's thesis on continual learning in enterprise LLMs — and serve as the real deployment environment for the 'Patch-and-Route' framework being researched.
1. Advanced RAG Server (Enterprise QM Chatbot)
A FastAPI application positioned between OpenWebUI and on-premise LLM backends (vLLM / llama.cpp), providing sophisticated document retrieval over Quality Management policies and procedures. All computation runs on local hardware — no data leaves the organization. Key capabilities:

Automatic data source routing (regex → embedding classification → anaphora history scan)
Hybrid retrieval combining dense semantic search and BM25, augmented by HyDE (Hypothetical Document Embeddings)
Multi-query expansion and cross-encoder reranking for precision
Abbreviation-aware retrieval with a managed enterprise glossary
Reference following, citation generation, and repetition prevention
Multi-turn anaphora resolution for coherent conversations
LoRA adapter fine-tuning support for continual knowledge updates
Deployed via Docker Swarm and HPC/SLURM with NVIDIA MIG GPU slices

2. Multi-Agent QA Generation Pipeline
An enterprise-grade pipeline that transforms Markdown QM documents into high-quality bilingual (DE/EN) fine-tuning datasets through a 5-agent chain:

Surveyor: Extracts evidence snippets and assigns intent categories
Analyst: Generates Chain-of-Thought reasoning and comprehensive answers (reasoning trace preserved in training data)
Persona: Rewrites questions through diverse user personas (e.g., 'The Skeptic', 'The Newbie') for dataset variety
Judge: First-pass quality control — approves, auto-corrects, or rejects pairs
Translator: Ensures bilingual DE/EN symmetry

Followed by a dual-layer validation stage: rule-based sanity checks, then DeepSeek R1 fact-checking for hallucination detection. Optimized for NVIDIA MIG GPU parallelization.
3. PDF Preprocessing Pipeline
A modular CLI tool for converting proprietary QM PDFs into clean, RAG-optimized Markdown using MinerU, with optional Vision LLM integration (Qwen3-VL / Mistral-Small) for image and diagram descriptions. Features parallel GPU conversion (~8x throughput via MIG slices) and an automated cleaning pipeline for TOC removal, abbreviation extraction, and YAML frontmatter generation.

Skills

Large Language ModelsRAGLoRA / PEFTMulti-Agent SystemsPythonFastAPIHPC / SLURMNVIDIA MIGFine-TuningNLP Research

September 2022 - January 2025

Full Stack Developer

Part Time

Fraunhofer IPK, Berlin, Germany

Back- and Frontend development using the MERN stack, maintenance and enhancement of the main software 'process assistant' (including integration of RAG chatbots), and agile collaboration within a Scrum team.

The Fraunhofer-Gesellschaft, headquartered in Germany, is one of the world’s leading organizations for applied research, playing a pivotal role in innovation by focusing on cutting-edge technologies and their transfer to industry.
At Fraunhofer IPK, I contributed to the development of digital solutions for integrated production as part of the Business Process and Factory Management department. My main responsibilities included:

Developing web applications using the MERN stack (MongoDB, Express.js, React.js, Node.js)
Designing and implementing RESTful APIs with Express.js
Creating user interfaces with React.js
Database design and management with MongoDB
Conducting code reviews and testing
Debugging and optimizing existing applications
Integrating third-party APIs

Additionally, I maintained and advanced the BatteryPass prototype, and developed an assessment tool to help companies prepare for the BatteryPass by identifying areas for improvement. I also led the integration of a RAG Chatbot that utilizes third-party chatbots like ChatGPT and Gemini within our process management tool, "Process Assistant," to facilitate document queries and element searches.
I identified a system bottleneck, improving backend runtime by 60-70% through caching implementation. Working in a team of 4-8 people, we conducted regular Scrum meetings. I held responsibility for maintaining the Process Assistant and led the BatteryPass and PA-assessment tool projects.

Skills

Full-Stack DevelopmentMERN StackSoftware EngineeringProject ManagementScrumCI/CD

January 2023 - April 2023

Bachelor Thesis

Full Time

Fraunhofer IPK, Berlin, Germany

Designed and developed the first prototype for the upcoming EU Digital Product Passport for traction batteries, set to launch in 2027.

For my bachelor's thesis, I designed and developed the first prototype for the upcoming EU Digital Product Passport for traction batteries, set to become mandatory in the EU in 2027. This project was conducted in the context of the '[Battery Pass](https://thebatterypass.eu/)' initiative, an EU-funded consortium aiming to establish a standardized digital passport for greater transparency and sustainability in the battery value chain, from raw materials to recycling.
The goal was to create a system providing verifiable data on production conditions, CO2 footprint, and circularity, accessible to consumers and stakeholders via a simple QR code scan. Key technical implementations included:

A track-and-trace environment built on a MongoDB database.
Use of a Fiware-Orion context broker for editable CRUD operations.
A backend function to calculate the CO2 footprint based on kWh.
The entire product pass was developed as a widget within Fraunhofer's 'Process Assistant' platform.

The functionality was tested and demonstrated using a Lego learning game to simulate the value chain. The project emphasized high usability and performance, for instance by using a hashmap for efficient search operations.

ComponentGrade

Thesis Gradevery good

Colloquium Gradegood

Skills

Software EngineeringDockerBPMN

June 2021 - August 2022

Working Student

Part Time

HTW Berlin, Germany

Core team member for the implementation of an improved campus management system.

As a working student in the university's executive office, I was part of the core team responsible for implementing a new Campus Management System (CaMS) at HTW Berlin. My main responsibilities included:

Providing administrative and organizational support to the project management, including organizing meetings and training sessions.
Visually preparing presentation materials.
Maintaining project-related communication channels such as the project homepage (using Typo3) and the project wiki (Confluence).
Contributing to test management, including test design, execution, and documentation from a student's perspective.
Advising on the selection of UI components (e.g., radio buttons, checkboxes) for the new web portal.
Assisting with process documentation.

Thanks to my strong IT skills, I became a valued contact person within the team and was able to effectively represent the students' requirements.

Skills

Web DevelopmentTypo3ConfluenceProject Management

Education

April 2023 - July 2026

M.Sc. in Computer Science

Humboldt University to Berlin, Germany

Focus: Natural Language Processing

GPA: 1.7 (current); projected 1.5 (~4.0 US) upon graduation

In my Master's program in Computer Science at Humboldt University of Berlin, I specialized in Natural Language Processing and focused my expertise in Data and Knowledge Management.
My coursework provided a comprehensive understanding of both theoretical foundations and practical applications. Key modules included:

Data and Knowledge Management:

Introduction to Natural Language Processing
Deep Learning for Visual Computing
Process Mining
Cognitive Robotics
Process Prediction and Machine Learning
Implementation of Databases
Statistics and Data Science

Model-Based System Development:

Business Process Automation
Secure and Reliable Software Systems
Psychology for Information Systems
Wireless Communication Systems

This curriculum was complemented by research-oriented seminars on 'Advanced Topics in Human-Computer Interaction' and 'Machine Learning in Robotics', as well as an advanced course in 'Efficient Machine Learning for Natural Language Processing'.

Skills

Natural Language ProcessingMachine LearningBusiness Process AutomationProcess MiningDeep Learning

February 2025 - June 2025

M.Sc. in Artificial Intelligence (Erasmus Exchange)

Universitat Politècnica de Catalunya in Barcelona, Spain

Focus: Advanced NLP, Machine Learning

During my Erasmus exchange semester at the Universitat Politècnica de Catalunya in Barcelona, I focused on advanced topics in Artificial Intelligence. This experience allowed me to deepen my understanding of cutting-edge concepts and gain international academic exposure. The courses I took there were:

Advanced Natural Language Processing
Human-Computer Interaction
Multi-Agent System Design
Professional Practice in Artificial Intelligence
Machine Learning
Spanish (Level A2)

Skills

Advanced NLPMulti-Agent System DesignDeep Learning

April 2020 - July 2023

B.Sc. in Computational Science and Engineering

HTW Berlin, Germany

Focus: Software Engineering and Embedded Systems

GPA: 1.5 (equivalent to ~4.0 GPA in the US system)

The Computational Science and Engineering program at HTW Berlin is an interdisciplinary course of study at the intersection of computer science, engineering, and mathematics. It provides a broad foundation in software development while integrating core principles from electrical and mechanical engineering.

Here is a complete list of my courses:

Mathematics
Algorithms
Introduction to Computer Science
Modeling in Computer Science
Programming
Database Systems
Technical Drawing and 2D-CAD
Design Principles and 3D-CAD
Construction
Thermodynamics
Fluid Mechanics
Technical Mechanics
Mechatronics
Introduction to Materials and Manufacturing Technology
Project Management
Scientific Work
Programming Project
Software Development Project
Interdisciplinary Project in Engineering Informatics
Embedded Systems
Mobile Applications in Engineering
Advanced Programming
Fundamentals of Business Administration
Legal Principles and Practices for Companies
1st Foreign Language: English

Skills

Database SystemsEmbedded SystemsMechatronicsSoftware EngineeringAlgorithms & Data StructuresProject Management

Voluntary Work

August 2019 - March 2021

Volunteer - Biodiversity Contact Person

Greenpeace, Mainz

Served as Biodiversity contact person for environmental initiatives.

See my projects