Student Theses

These master thesis topics are currently open and available for students pursuing a degree in BIS and, in some cases, for students in the Master of Medical Informatics programs at FHNW.

RepoChat Related Topics

These thesis topics are part of the RepoChat Project (Semantic Verification in Large Language Model-based Retrieval Augmented Generation), funded by Innosuisse under grant 109.093 IP-ICT. The project aims to enhance the accuracy and relevance of AI-generated responses by improving semantic verification in large language models.

  • Martin, A., Witschel, H. F., Mandl, M., & Stockhecke, M. (2024). Semantic Verification in Large Language Model-based Retrieval Augmented Generation. Proceedings of the AAAI Symposium Series, 3(1), Article 1. https://doi.org/10.1609/aaaiss.v3i1.31199

  • Martin, A., Witschel, H. F., Mandl, M., & Stockhecke, M. (2024, März 26). Semantic Verification in Large Language Model-based Retrieval Augmented Generation. AAAI Spring Symposium on Empowering Machine Learning and Large Language Models with Domain and Commonsense Knowledge (AAAI-MAKE), Stanford University, California, USA. Zenodo. https://doi.org/10.5281/zenodo.10892026

Human-in-the-Loop Prompt Tuning for Knowledge Graph Generation

Teaser: Explore prompt tuning methodologies augmented by continuous human feedback to enhance the semantic accuracy and contextual relevance of knowledge graph generation by large language models (LLMs).

Business Impact: Improves the reliability, precision, and efficiency of automated knowledge graph construction, significantly reducing the need for manual intervention while enhancing overall data quality.

Research Fields/Areas: Artificial Intelligence, LLMs, Prompt Tuning, Knowledge Graphs, Human-in-the-Loop Systems

Subject Description: This thesis investigates prompt tuning processes enhanced by iterative human feedback, with the aim of improving the quality and semantic accuracy of knowledge graph generation performed by LLM-based agents. The research focuses on integrating human-in-the-loop mechanisms into prompt adaptation workflows, enabling dynamic refinement based on contextual and domain-specific input. The project is part of the RepoChat Project, which seeks to advance semantic verification in retrieval-augmented generation (RAG) systems. By systematically combining human insight with automated prompt optimization, this thesis contributes to the development of more robust and precise knowledge graph construction pipelines.

Fine-Tuning and Alignment of Large Language Models Using Human Feedback for Knowledge Graph Generation

Teaser: Investigate advanced fine-tuning and alignment techniques that incorporate structured human feedback to improve the semantic fidelity and contextual reliability of knowledge graphs generated by LLMs.

Business Impact: Enhances the quality and trustworthiness of AI-generated knowledge graphs, enabling better-informed decision-making and reducing the resource burden of manual post-processing.

Research Fields/Areas: Artificial Intelligence, LLMs, Fine-Tuning, Knowledge Graphs, Human Feedback Alignment

Subject Description: This thesis explores the post-training fine-tuning and alignment of large language models through the integration of human expert feedback. The goal is to increase the semantic precision and contextual appropriateness of LLM-generated knowledge graphs. This research contributes to the RepoChat Project, which focuses on improving semantic verification within retrieval-augmented generation (RAG) environments. By aligning model behavior with expert knowledge through targeted feedback loops, the thesis aims to advance methods for trustworthy and application-ready AI-generated knowledge representations.

Hybrid Multi-Agent Semantic Verification Combining Knowledge Graph Matching and LLM Agents

Teaser: Design a hybrid semantic verification framework that combines explicit knowledge graph matching with agent-based validation by large language models to ensure semantic consistency and contextual correctness.

Business Impact: Increases the reliability and robustness of AI-generated outputs by integrating structured and unstructured reasoning in semantic verification workflows.

Research Fields/Areas: Artificial Intelligence, Multi-Agent Systems, Knowledge Graphs, LLMs, Semantic Verification

Subject Description: This thesis develops a hybrid semantic verification approach based on the coordination of multiple agents. The framework integrates explicit knowledge graph matching techniques with the interpretive capacities of LLM-based agents. Situated within the RepoChat Project, the system is designed to validate outputs in RAG settings by leveraging both symbolic representations and language-based inference. The result is a verification architecture that improves the trustworthiness and contextual adequacy of AI-generated responses in complex information environments.

HIVBOT Related Topics

These thesis topics are part of the international SNSF-funded HIVBOT Project (Researching Intelligent Chatbots as Healthcare Coaches), which aims to support people living with HIV. The project focuses on developing intelligent chatbots to provide reliable, safe, and accessible healthcare information and support.

  • Martin, A., Pande, C., Schwander, S., Ajuwon, A. J., & Pimmer, C. (2024). Domain-specific Embeddings for Question-Answering Systems: FAQs for Health Coaching. Proceedings of the AAAI Symposium Series, 3(1), Article 1. https://doi.org/10.1609/aaaiss.v3i1.31197

  • Pande, C., Martin, A., & Pimmer, C. (2023). Towards Hybrid Dialog Management Strategies for a Health Coach Chatbot. In A. Martin, H.-G. Fill, A. Gerber, K. Hinkelmann, D. Lenat, R. Stolle, & F. van Harmelen (Eds.), Proceedings of the AAAI 2023 Spring Symposium on Challenges Requiring the Combination of Machine Learning and Knowledge Engineering (AAAI-MAKE 2023). CEUR-WS.org. http://ceur-ws.org/Vol-3433

  • Martin, A. (2024). Ensuring Trustworthy Dialogue Systems. Zenodo. https://doi.org/10.5281/zenodo.11530754

Agentic FAQ: Trustworthy Response Management and Confirmatory Dialogue in Healthcare Applications

Teaser: Extend a healthcare-focused FAQ system with structured response guardrails and confirmatory questioning to improve the safety, accuracy, and clarity of non-critical automated interactions.

Business Impact: Strengthens patient communication by delivering consistent and contextually appropriate responses, thereby reducing miscommunication and improving trust in AI-powered services.

Research Fields/Areas: Artificial Intelligence, Dialogue Systems, LLMs, NLP, Healthcare Informatics, Human-AI Interaction

Subject Description: This thesis focuses on enhancing a conversational FAQ agent used in healthcare by integrating response guardrails and confirmatory dialogue techniques for handling non-critical inquiries. The research targets the design of a safe and structured response framework aimed at improving user comprehension, trust, and satisfaction. The project is part of the SNSF-funded HIVBOT initiative and aspires to produce publishable methodological contributions to the field of human-AI healthcare communication.

Low-Threshold Hybrid Dialogue Framework with Enhanced Privacy and Messaging Integration for Public Health

Teaser: Develop a low-cost, table-driven dialogue system with built-in privacy safeguards, scheduling features, and integration with messaging platforms such as WhatsApp for scalable deployment in public health scenarios.

Business Impact: Enables secure and scalable conversational systems for resource-constrained environments, enhancing public health communication and patient engagement.

Research Fields/Areas: Artificial Intelligence, Dialogue Systems, LLMs, Privacy Engineering, Health Informatics, Mobile Communication Technologies

Subject Description: This thesis designs a privacy-aware, hybrid dialogue framework optimized for public health use cases. The system combines a table-driven logic layer with secure data handling, reminder features, and real-time integration with messaging platforms like WhatsApp. Drawing on insights from both educational technology development and the HIVBOT Project, the thesis targets usability, accessibility, and operational efficiency. The resulting system supports scalable, secure communication infrastructures suited to low-resource environments.

Emergency Situation Detection in AI-Based Public Health Dialogue Systems

Teaser: Develop intelligent detection mechanisms for identifying high-risk mental health emergencies, such as suicidal ideation, in AI-driven healthcare dialogue systems.

Business Impact: Improves patient safety through early recognition and timely escalation of critical mental health cases to appropriate support structures.

Research Fields/Areas: Artificial Intelligence, LLMs, NLP, Emergency Detection, Mental Health Informatics, Conversational Safety Systems

Subject Description: This thesis explores methods for detecting mental health emergencies in conversational agents, focusing on the identification of crises such as suicidal episodes. Leveraging advanced LLM and NLP techniques, the system aims to recognize linguistic and contextual indicators of distress. The project includes the design of a referral protocol that links at-risk users to emergency services or professional support. Conducted within the SNSF-funded HIVBOT framework, this work advances AI safety in sensitive public health applications.

Further Agentic AI and Assistant Coding Topics

Integrating Vibe and Agentic Coding into Business and Digital Education: A Framework for Competency-Driven Learning

Teaser: Examine the evolving skill requirements in business education and develop a competency-based instructional model that leverages Vibe and Agentic Coding to foster critical, analytical, and computational thinking.

Business Impact: Equips future professionals with skills aligned with AI-centric work environments, enhancing adaptability and strategic thinking in dynamic decision-making contexts.

Research Fields/Areas: Business Education, Critical Thinking, Computational Thinking, AI-Enhanced Learning, Educational Design

Subject Description: This thesis investigates how AI-supported coding paradigms—specifically Vibe Coding and Agentic Coding—can be integrated into business education to promote core cognitive competencies. Drawing on Singh, Guan, and Rieh’s (2025) findings on the effectiveness of metacognitive prompts in promoting critical thinking during AI-supported search (arXiv:2505.24014), the research designs and evaluates learning interventions that use prompt-based reflection and agentic reasoning tasks. The goal is to operationalize a framework for business and digital education that builds metacognitive awareness and future-oriented problem-solving skills.

Reference: Singh, A., Guan, Z., & Rieh, S. Y. (2025). Enhancing Critical Thinking in Generative AI Search with Metacognitive Prompts. arXiv:2505.24014. https://doi.org/10.48550/arXiv.2505.24014

Conversational BPMN Service Integration: Democratizing Workflow Automation via Vibe and Agentic Coding

Teaser: Create a conversational framework for BPMN service task integration using Vibe and Agentic Coding, enabling non-technical users to orchestrate external services via External Task Workers (Camunda) or Job Workers (Flowable).

Business Impact: Reduces technical barriers to process automation, enabling broader stakeholder participation and increasing agility in process-centric organizations.

Research Fields/Areas: Business Process Management, Workflow Automation, Conversational AI, Agentic Systems, Low-Code Tools

Subject Description: This thesis explores a novel approach to BPMN-based process integration that combines conversational interfaces and agentic reasoning. Targeting workflow engines like Camunda and Flowable, the system uses Vibe Coding to allow natural language description of integration logic and Agentic Coding to autonomously execute service orchestration. External Task Workers (Camunda) and Job Workers (Flowable) are used to implement the operational backend. The research includes the design of a prototype and its evaluation in terms of usability, scalability, and democratization of automation for non-developer users.

Last updated