Experience

 
 
 
 
 
April 2021 – Present
Renningen, Germany

Research Engineer

Bosch Corporate Research, Robert Bosch GmbH

Developing AI, knowledge, and data engineering solutions across various Bosch Business Units. Key projects include:

  • AI Reliability: Developing tools to formally verify AI-generated code fixes and architecting intelligent context-provisioning for LLMs.
  • Semantic Data Integration: Built a cross-system "global search" platform powered by knowledge graphs and managed evolving ECU software label equivalence.
  • Data Architecture: Designed semantic data models for the Nevonex+ project within the Agri-Gaia/Gaia-X ecosystem.
 
 
 
 
 
November 2013 – April 2021
Koblenz, Germany

Researcher, Akademischer Rat

Institute for Web Science and Technologies, University of Koblenz-Landau

Akademischer Rat since April 2020. Conducted research at the intersection of programming language theory and semantic web technologies. Key contributions include:

  • Type Systems for Programming with Semantic Graphs: Authored a Ph.D. thesis exploring formal type systems to prevent runtime errors in programs querying and processing graph data.
  • Graph Data Querying: Advanced the integration of query languages into programming environments, including supervising theses on these integrations, on mapping SPARQL to SQL for virtual knowledge graphs. Also worked in the area of data integration (Ontology-based data integration) with graph databases and working with graph databases in general.
  • Academic Instruction: Lectured and designed coursework for courses Algorithms and Datastructures, Artificial Intelligence, and Big Data.
 
 
 
 
 
August 2009 – October 2009
Mainz, Germany

Intern at TPC Development

IBM Deutschland Research & Development GmbH

Supported the development of Tivoli Productivity Center by setting up and configuring a virtualized SoFS clustered NAS environment.

Selected Projects

ForML-S (AI Reliability

Bosch Corporate Resarch | Jan. 2024 - Present

  • Objective: Enable the safe deployment of Generative AI for source code within safety-critical engineering domains.
  • Approach: Through a continous feedback loop integrating Large Language Models (LLMs) with formal source code verification tools, including model checking and fuzzing, to execute automated code transformations.
  • Applications: Powering automated translation between programming languages and generating reliable, semantic-preserving fixes for code alarms and design smells.
More information available in the Bosch Research Blog.

Global Search @ PS-EC

Bosch Corporate Research | Apr. 2021 - Dec. 2023

  • Objective: Streamline the discoverability of automotive data (specifically ECU calibration data) to make it ready for downstream AI tasks.
  • Approach: Engineered a unified knowledge graph spanning multiple heterogenous data sources to create a highly searchable interconnected layer.

Nevonex+ (Agri-Gaia/Gaia-X Ecosystem)

Bosch Corporate Research | Jan. 2021 - Dec. 2023

  • Objective: Built an enterprise data platform to ingest, store and process telemetric data from agricultural machinery and edge sensors. Provided the data to downstream AI tasks and integrated it into the Agri Gaia dataspace.
  • My Role: Served as the data architect, owning the end-to-end data ingestion pipeline, semantic modeling and integration architecture.
  • AI and Visualization: Developed a prototypical AI model to function as a virtual machine sensor and built data visualization dashboards to highlight how actionable agricultural insights can be unlocked.
More information can be found on the Bosch Homepage.

LISeQ (Language Integrated Semantic Queries)

University of Koblenz | Sep. 2018 - Aug. 2021

  • Objective: To eliminate runtime errors caused by structural mismatches when embedding semantic queries into strongly typed programming languages.
  • Approach: Researched and developed formal type-checking mechanisms that evaluate embedded SPARQL queries at compile time.

Selected Publications

All Publications »

, , , ,
Proceedings of the 19th International Semantic Web Conference (ISWC)

The Shapes Constraint Language (SHACL) allows for formalizing constraints over RDF data graphs. A shape groups a set of constraints that may be fulfilled by nodes in the RDF graph. We investigate the problem of containment between SHACL shapes. One shape is contained in a second shape if every graph node meeting the constraints of the first shape also meets the constraints of the second. To decide shape containment, we map SHACL shape graphs into description logic axioms such that shape containment can be answered by description logic reasoning. We identify several, increasingly tight syntactic restrictions of SHACL for which this approach becomes sound and complete.
PDF DOI
, , , ,
Proceedings of the 18th International Semantic Web Conference (ISWC)

It is a strength of graph-based data formats, like RDF, that they are very flexible with representing data. To avoid run-time errors, program code that processes highly-flexible data representations exhibits the difficulty that it must always include the most general case, in which attributes might be set-valued or possibly not available. The Shapes Constraint Language (SHACL) has been devised to enforce constraints on otherwise random data structures. We present our approach, Type checking using SHACL (TyCuS), for type checking code that queries RDF data graphs validated by a SHACL shape graph. To this end, we derive SHACL shapes from queries and integrate data shapes and query shapes as types into a ${\lambda}$-calculus. We provide the formal underpinnings and a proof of type safety for TyCuS. A programmer can use our method in order to process RDF data with simplified, type checked code that will not encounter run-time errors (with usual exceptions as type checking cannot prevent accessing empty lists).
PDF DOI
, ,
Proceedings of the 26th European Symposium on Programming (ESOP)

Semantic data fuels many different applications, but is still lacking proper integration into programming languages. Untyped access is error-prone. Mapping approaches cannot fully capture the conceptualization of semantic data. In this paper, we present ${\lambda_{DL}}$, a typed ${\lambda}$-calculus with constructs for operating on semantic data. This is achieved by the integration of description logics into the ${\lambda}$-calculus for both typing and data access or querying. The language is centered around several key design principles, in particular: (1) the usage of semantic conceptualizations as types, (2) subtype inference for these types, and (3) type-checked query access to the data by both ensuring the satisfiability of queries as well as typing query results precisely. The paper motivates the use of a designated type system for semantic data and it provides the theoretic foundation for the integration of description logics as well as the core formal definition of ${\lambda_{DL}}$ including a proof of type safety.
PDF DOI
, , , , ,
Proceedings of the 13th International Semantic Web Conference (ISWC)

The Semantic Web is intended as a web of machine readable data where every data source can be the data provider for different kinds of applications. However, due to a lack of support it is still cumbersome to work with RDF data in modern, object-oriented programming languages, in particular if the data source is only available through a SPARQL endpoint without further documentation or published schema information. In this setting, it is desirable to have an integrated tool-chain that helps to understand the data source during development and supports the developer in the creation of persistent data objects. To tackle these issues, weintroduce LITEQ, a paradigm for integrating RDF data sources intoprogramming languages and strongly typing the data. Additionally, we report on two use cases and show that compared to existing approaches LITEQ performs competitively according to the Halstead metric.
PDF DOI

Teaching

During my time at the University of Koblenz, I was responsible for designing, lecturing, and leading tutorial for both undergraduate and graduate courses, algonside supervising specialized research seminars.

Algorithms and Data structures

An undergraduate course focusing on the design and complexity analysis of foundational algorithms and data structures. Core topics include:

  • Searching and sorting algorithms
  • dynamic data structures (lists, trees, hashing, graphs)
  • graph algorithms (shortest path, minimal spanning trees, maximum flow)

Artificial Intelligence

A graduate-level course focusing on classical, symbolic AI, with an emphasis on logic, automated planning, and state-space search strategies. Core topics include:

  • Propositional and first-order logic
  • declarative pogramming in Prolog
  • planning and searching in state space
  • Non-monotonic logic (Default Logic, Answer Set Programming (ASP), and Abstract Argumentation
  • Agent-based systems

Big Data

A graduate-level course covering modern distributed systems, large-scale data processing architectures, and analytical database models. Core topics include:

  • Cloud computing paradigms
  • distributed programming with Apache Spark
  • Online Analytical Processing (OLAP)
  • and NoSQL databases architectures

Other teaching activities

Beyond regular curriculum courses, I regularly led hands-on research practicals. Notable examples include specialized seminars on graph algorithms and a fun research practical focused on developing AI agents for Starcraft.

Contact