Open Topics

Open thesis and research collaboration topics combining Knowledge Graphs, Retrieval-Augmented Generation, prompt engineering, time-series imputation, and data sovereignty for predictive maintenance in connected vehicles. These open topics are designed to inspire new research directions, foster collaboration, and drive innovation at the intersection of LLMs, causal AI, and vehicle telematics. Each topic includes a brief explanation and key research questions to guide exploration and experimentation.

01

Knowledge Graph-Augmented Prompting for Explainable Vehicle Telematics Analytics

Brief explanation

This topic combines LLM prompting with a Knowledge Graph containing vehicle components, telematics variables, and causal relations to improve explainability.

Research Questions

  • How can a Knowledge Graph improve the reliability of LLM-generated explanations in vehicle telematics analysis?
  • Does KG-augmented prompting reduce hallucinated causal relationships compared to standard prompting?
  • How well do KG-supported LLM outputs align with domain-informed causal ground truth?
02

Multi-Criteria Prompt Selection for LLM-Based Causal Reasoning in Predictive Maintenance

Brief explanation

This topic extends the idea of multi-criteria decision analysis from imputation selection to prompt selection. Prompts would be ranked based on causal accuracy, consistency, leakage risk, and explanation quality.

Research Questions

  • How can multi-criteria decision analysis be used to select the best prompting strategy for causal reasoning?
  • Which evaluation criteria are most important for selecting prompts in predictive maintenance use cases?
  • Does multi-criteria prompt selection improve LLM reliability compared to choosing prompts based only on accuracy?
03

Effect of Time-Series Imputation Quality on LLM-Based Causal Reasoning

Brief explanation

This topic connects Time-Series Modelling, Time-Series Evaluation, and Time-Series with LLM Modelling. It investigates whether better imputation performance leads to better LLM causal reasoning.

Research Questions

  • How does imputation quality influence LLM-generated causal graphs?
  • Do lower RMSE and higher R² in imputation results lead to better causal graph precision, recall, and F1 score?
  • Which imputation paradigm, statistical, ML, or DL provides the most reliable input for LLM-based causal reasoning?
04

Ground Truth Validation of LLM-Based Causal Graphs Using Imputed Time-Series Data

Brief explanation

This topic is closely related to the Causal analysis "Ground Truth Validation" based on the real-world data-driven vehicle telematics analysis. It focuses on validating LLM-generated causal graphs against ground truth created from literature, domain knowledge, and dataset-derived patterns.

Research Questions

  • How can a reliable reference graph be constructed for evaluating LLM-generated causal graphs?
  • How do imputation methods affect the agreement between LLM-generated graphs and ground truth?
  • Which structural metrics best capture the quality of LLM-generated causal graphs?
05

RAG and Knowledge Graph Prompting for Explainable Predictive Maintenance Decisions

Brief explanation

This topic combines LLM Modelling, Knowledge Graph utilization, and Predictive Maintenance Types such as Anomaly Detection, State of Health, and Remaining Useful Life.

Research Questions

  • How can RAG and Knowledge Graphs improve the factual grounding of LLM-based predictive maintenance explanations?
  • Does retrieved domain knowledge improve the quality of causal explanations generated by LLMs?
  • Can RAG-KG prompting reduce hallucinated maintenance recommendations in vehicle telematics analysis?
06

Data-Sovereign Causal AI Pipeline for V2X Predictive Maintenance

Brief explanation

This topic connects Data Space, V2X Architecture, Security, Time-Series Modelling, LLM Modelling, and Causal AI. It designs an end-to-end architecture for predictive maintenance under data sovereignty constraints.

Research Questions

  • How can vehicle telematics data be processed in a data-sovereign pipeline for causal AI-based predictive maintenance?
  • How can V2X architecture support secure data collection, imputation, LLM reasoning, and causal discovery?
  • What role do Knowledge Graphs and local LLM deployment play in improving explainability, sovereignty, and trust?