Lulienne Schiefelbein

During my corporate student tenure at Accenture, I completed several internships focusing on the Consumer Goods Industry and Salesforce CRM solutions. My roles encompassed business analysis, data import, development artifact demos, product management support, Salesforce Wave Analytics and Report Builder PoCs, SOQL and APEX skills acquisition, Python-based ETL processes, and the examination of visualization tools. In my bachelor thesis, I applied machine learning to predict out-of- stocks in retail stores.

I’ve led the Java backend development of the electronic health record (elektronische Patientenakte, ePA) as a product owner and solution architect at IBM. This mobile application caters to millions, including medical practitioners, insurance companies, and insured individuals. My responsibilities encompassed requirement gathering, software development process enhancement through DevSecOps and Secure Software Development Lifecycle (SDLC) practices, and mentoring junior team members. I’ve also actively contributed to architectural design, addressed production issues, and ensured regulatory compliance. My proficiency in Python allowed me to automate requirements engineering processes and provide innovative solutions.

As a Cognitive Consultant, I contributed to a project involving the extraction of IFRS16 relevant data from contracts in multiple European languages and formats. BigInsights was applied for OCR followed by Named Entity Recognition (NER) using traditional rule- based approaches and machine learning based models. In my role I annotated contract data and trained a machine learning model in Watson Knowledge Studio, which enabled automatic Named Entity Recognition of contract values, streamlining operations and reducing errors. Additionally, I voluntarily supported a project focused on extracting named entities from customer complaints using Python pipelines and Watson Knowledge Studio.

I played a crucial role in a project aimed at transforming unstructured cooking recipes into a structured format, enabling efficient machine-based processing. This initiative sought to enhance recipe analysis, simplifying tasks like recipe comparison and component substitution through innovative graph-based visualization. My primary focus revolved around training and evaluating underlying models, establishing robust pipelines using pandas, Spacy, Huggingface, and PyTorch for Named Entity Recognition (NER). I also developed rule-based relationship extraction methods and utilized the GraphViz library for graph visualizations. Seaborn has been used to visualize the evaluation results and inter-annotator agreement scores for a jointly labeled corpus of 200 recipes. We created a user-friendly interface with Streamlit. Notably, the project achieved remarkable success with F1 scores exceeding 90% for models such as BERT, RoBERTa, and MPNET. We also implemented the Nunamaker research framework, ensuring systematic project execution and broader applicability across domains.

My master’s thesis centered on automating patent similarity assessment, a critical aspect of innovation management and intellectual property analysis. Notably, I collaborated closely with the Fraunhofer Institute IAO in Stuttgart throughout this project. I compiled a substantial dataset of 92,000 US patents from Google BigQuery and crafted a dataset featuring 50,000 similar patent pairs, leveraging the “X” citation relationship. Additionally, I curated expert datasets to facilitate cross-domain assessment. The project’s core involved the fine-tuning of eight machine learning models, employing the sentence-transformers framework fine-tuned to capturing semantic connections within patent texts. Furthermore, I showcased the practical utility of fine-tuned models by encoding patent textual features into vector embeddings for visualization and similarity computations. Presently, I continue to contribute to this project by transforming my work and that of fellow researchers into an invaluable asset for patent analysis.