Data Scientist/Researcher in research project Alisa34 ? Automation of regulatory processes with AI (2025, University of Applied Science of South Westphalia)

• Development of an AI-supported framework for the semi-automated processing of authorisation processes in the field of nuclear decommissioning. With the help of open-source large language models (LLMs), neurosymbolic modelling and agent-based systems, regulatory documents are to be evaluated more efficiently, knowledge mapped consistently and decisions supported in a well-founded manner. The project is being implemented together with an industry partner, which ensures direct practical relevance and the subsequent application of the developed solutions in real approval processes.

My Contribution to Success:

Document processing:

• Extraction and structuring of regulatory content (e.g. laws, safety assessments, technical instructions) from heterogeneous formats (PDF, Word, HTML, scanned documents) using OCR, document layout detection and classic NLP techniques. 

Semantic search and retrieval:

• Development of a retrieval framework for semantic document search and knowledge linking (incl. ontology connection), using classic RAG approaches and GraphRAG methods 

Training of LLMs:

• Domain-specific fine-tuning of LLMs for better indexing of German-language regulatory texts and for context-sensitive answers to questions in the authorisation process.

• Development and evaluation of GraphRAG approaches to improve knowledge access and decision support in the field of nuclear decommissioning. The aim was to make regulatory, legal and safetyrelated documents more efficiently accessible - while complying with high data protection requirements and using purely local open source models. The focus was on modelling document contexts, ontologies and retrieval strategies in a safety-critical, knowledge-intensive field of application. 

My Contribution to Success

• Design, implementation and comparison of two GraphRAG architectures with classic RAG pipelines to improve semantic retrieval in regulatory domains.
• Development of a locally running, data protection-compliant RAG infrastructure based on open source LLMs including PDF parsing, OCR, chunking, keyword normalisation and Neo4j graph modelling

Results and Benefits

• The work shows how GraphRAG methods can improve knowledge access in highly complex and regulated domains - with a focus on transparency, explainability and data protection. The methods developed are transferable to other safety-critical industries (e.g. pharmaceuticals, energy, aviation) and are incorporated into current industry and research projects, in particular for the further development of RAG systems for German-speaking specialised domains.

Implementation of an MES system for digitalisation of production processes. Development and support of the interfaces

• As part of the project, a Manufacturing Execution System (MES) was implemented to digitalise and optimise production processes and create end-to-end transparency in real time. The aim was to ensure seamless communication between production facilities, IT systems and the new MES in order to utilise production data more efficiently and automate processes.
• The main task was to develop and support the interfaces that linked the MES with existing systems such as SAP and the production facilities via OPC-UA.

My Contribution to Success

Interface development:

• Implementation of OPC-UA interfaces for connecting production systems to the MES, integration with SAP for bidirectional data transfer and SQL database connections.

Technical support:

• Monitoring, optimisation and troubleshooting of the interfaces during the project and operational phase, ensuring performance and availability.

Cooperation:

• Close coordination with team members and external partners, support with tests and user training in handling the interfaces.

Results and Benefits

Optimised production processes: 

• By integrating the MES, processes could be organised more efficiently and monitored in real time.

Seamless system communication:

• The interfaces developed enabled smooth data transfer between machines, IT systems and the MES.

Future-proof architecture:

• The solution is flexibly scalable and offers a stable basis for further digitalisation measures.

• The aim of the project was to demonstrate the potential and use cases of Retrieval Augmented Generation (RAG) and Large Language Models (LLMs) to the company. A proof of concept (PoC) was to show how these technologies can be used profitably to increase efficiency and added value in various business areas. 

My Contribution to Success

Consulting and conception:

• Analysis of business requirements, consulting on the selection and use of technologies such as Langchain, Huggingface Transformers and Vector & Graph Databases, as well as support with the architecture concept on Azure.

Coordination and implementation:

• Close cooperation with the data scientist on PoC implementation and advice on best practices in the use of RAG and LLMs.

Results and Benefits

Proof of value creation:

• The PoC successfully demonstrated that RAGs and LLMs can deliver significant efficiency and insight gains for business processes.

Technological basis:

• Development of a scalable architecture that enables the use of LLMs and RAGs to be extended to other areas of the company.

Strategic consulting:

• Clear recommendations for the long-term use of the technologies, customised to the individual requirements of the company

Development of a digital shift book to optimise communication at production level

• As part of the digitalisation of production, a digital shift book was developed that replaces manual processes with a paperless solution and significantly improves communication and the flow of information at production level. The aim of the project was to implement a centralised, intuitive platform for recording and transferring shift information that is directly integrated into existing IT and OT systems.
• The project was of crucial importance to promote efficiency, traceability and transparency in the production processes.

My Contribution to Success

Interface development:

• Implementation of REST interfaces to integrate the existing MES and OPC-UA for connecting production machines for real-time data.

System integration: 

• Ensuring seamless integration into the IT and OT infrastructure in close cooperation with the development of the SAP UI5 interface and SAP customisation.

Requirements management:

• Translation of production requirements into technical solutions through regular coordination with the specialist departments.

Results and Benefits

Efficient communication:

• Significantly improved shift handovers and optimised information flow at production level.

Internal demand:

• Success led to interest from other production companies within the group.

Paperless processes:

• Replaced manual documentation, reduced errors and labour.

• Consultancy and implementation of IT solutions, mainly in the production environment. From evaluation of feasibility to finished solutions.

• The aim of the project was to develop proof-of-concepts (PoCs) that translate current research findings on large language models (LLMs) and retrieval augmented generation (RAG) into practical applications. The PoCs served as presentation and test platforms to demonstrate the performance of modern AI technologies and secure funding and corporate support.
• A particular focus was on the use of open-source models and the customisation of the systems for outof-domain use cases that deviated from the data in typical  researchpublications.

My Contribution to Success

Development of PoCs:

• Creation of interactive AI demos with the Langchain tech stack and Streamlit combined with retrieval mechanisms via vector and graph databases, as well as the use of LLMs. Implementation of PDF parsing to make unstructured data usable.

Results and Benefits

Funding and cooperation:

• PoCs played a key role in securing funding totalling EUR 4 million and gaining the support of other companies.

Innovative solutions:

• Successful adaptation of LLMs and RAG approaches to ?out-of-domain? data, expanding the range of possible applications.

Interdisciplinary collaboration:

• Successful coordination with researchers and scientific leaders to develop innovative PoCs and demonstrate their potential.

• Development of RAG applications and training of LLMs. 
• Research contribution to RAG and LLM. 
• Successful contribution to the acquisition of millions in funding.

Acquisition and evaluation of machine and process data for process optimisation

• The aim of the project was to develop a system for systematically recording and analysing machine and process data in order to make production processes and machine commissioning more efficient. Databased insights were used to uncover optimisation potential and create an improved understanding of complex production processes.

My Contribution to Success

Requirements management and coordination:

• Coordination with stakeholders to define objectives and organise collaboration in an interdisciplinary team.

Technical implementation:

• Development of a solution for real-time data acquisition with OPC-UA, storage in an SQL database and visualisation of the results with Microsoft PowerBI.

Process optimisation:

• Analysis of the data to identify potential improvements for production and commissioning processes together with process optimisers.

Communication:

• Presentation of the results and recommendations in an understandable form for technical teams and management.

Results and Benefits

Improved process understanding:

• The system created deeper insights into production processes and supported data-driven decisions.

Increased efficiency during commissioning:

• The commissioning of complex machines was made significantly faster and less error-prone.

Action-orientated dashboards:

• The visualisation of data enabled a quick understanding  and facilitated the implementation of optimisations.

• As part of a strategically important project, an integration platform was successfully migrated from version 3 to version 4 and a redundancy system was implemented at the same time in order to sustainably optimise availability and reliability. The platform is a central component of the company's digitalisation strategy and supports the implementation of Industry 4.0 initiatives, particularly in the connection between IT and OT systems.
• The project was characterised by a high level of complexity, as the new platform version revealed technical weaknesses during implementation. These challenges were mastered with ease thanks to practical solutions and a consistent focus on the project objectives.

My Contribution to Success

Project management:

• Planning, control and implementation of the project with responsibility for time and resource management and on-time delivery.

Technical implementation:

• Migration and integration of the platform with IT and OT systems using C#, OPC-UA, ABAP, SQL, MongoDB and REST, implementation a redundancy system for high availability.

Problem-solving expertise:

• Creative development of customised solutions for the challenges of the new platform version; development of a monitoring system with Grafana and InfluxDB.

Communication:

• Coordination with specialist departments (production, controlling, facility management, etc.) and experts from key technologies; transparent reporting to stakeholders.
  

Results and Benefits

Increased system availability:

• Redundancy system raises stability to a new level.

Digitalisation:

• platform as the basis for Industry 4.0 applications.

Efficient implementation:

• On-time target achievement despite technical challenges.

• As part of this project, a retrieval augmented generation (RAG) system was developed that can efficiently process and provide time-sensitive information in the context of the German Football League. The aim was to implement a powerful RAG system that combines both classic keyword-based approaches and modern methods such as time-aware retrieving and semantic search technologies.

My Contribution to Success

Hybrid RAG:

• Implementation of a hybrid RAG with a time-sensitive component that evaluates retrieval results based on the time information in the query and the creation time of the documents. A keyword search and a vector retrieval and a workflow with huggingface transformers and Langchain were integrated for this purpose. 

NLP and NER:

• Use of Spacy and NLTK to identify football-specific entities and time-based contextualisation.

Prompt Engineering:

• Development of precise prompts for relevant and accurate answers.

Results and Benefits

Powerful RAG system:

• The developed system combines semantic, keyword and time-based search methods and provides relevant, time-sensitive answers.

Real-time functionality:

• By optimising the retrieval and generation processes, a high level of efficiency and accuracy has been achieved that is suitable for real-time applications.

Flexibility and scalability:

• The modular architecture allows easy customisation to other sports or dataintensive use cases.

• As part of this university project, a system was developed to analyse job advertisements from various job portals. The aim was to gain valuable insights into the most frequently required skills in certain occupational groups and to analyse the salary structures of different sectors and companies. These analyses should help to facilitate targeted further training measures and create transparency about salary structures.
• The core of the project involved collecting data using a web crawler and processing and analysing this data using modern NLP methods and language models.

My Contribution to Success

Web crawler development:

• Automated collection and normalisation of relevant data such as skills, salaries and industry information using Scrapy and BeautifulSoup.

Data analysis:

• Cleansing and transformation of data with Pandas; extraction of key terms with SpaCy, NLTK and with the help of language models.

Visualising results: 

• Creating reports and visualisations to present findings and possible applications.

Results and Benefits

Targeted further training:

• Analysis revealed in-demand skills and qualification gaps.

Transparency in salaries:

• Sound salary benchmarks of occupations across industries and companies.

Scalability:

• Flexible system for additional data sources and specific analyses.

My Contribution to Success

Software development and architecture:

• Development of a modular, scalable and secure software solution in C#, including communication interfaces for MES and third-party systems as well as implementation of cryptography for data integrity.

FDA compliance:

• Implementation of the requirements in accordance with 21 CFR Part 11 with audit trails, electronic signatures and audit-proof traceability.

Requirements management:

• Coordination with stakeholders to record and validate regulatory and technical requirements and translate them into a suitable software solution.

Results and Benefits

New sales potential: 

• The software solution led to the development of a new business area that generates significant sales. 

FDA compliance:

• Successful development of legally compliant software that fulfils the strict requirements of the pharmaceutical industry.

Efficient integration:

• The solution enables secure and efficient communication between machines and third-party systems such as MES and ERP systems.