large language models

AI-Driven Support For Military Training Documentation Design

Military documentation plays a central role in training and operational practice, yet its quality is rarely evaluated from the perspective of actual use. Documents tend to prioritise completeness and regulatory compliance over clarity. This results in materials that are information dense, cognitively demanding, and pro...

From Text to Neural Embeddings: Structuring Conceptual Knowledge in Large Language Models

The broad adoption of Large Language Models (LLMs) has transformed the landscape of Natural Language Processing (NLP) since the early 2020s, placing this PhD thesis within one of the fastest-growing areas of recent years. This research project first investigates the capabilities of language models for the NLP task of I...

AI-Based Methods for Reliable Building Performance Management

Reliable building performance management involves reducing energy use and peak demand while maintaining indoor comfort and operational safety. Achieving these objectives requires effective strategies across the building life cycle, from early-stage design to operation. However, the availability, quality, and structure...

EcoxPU: Autonomous Multi-Agent Orchestration for Energy-Efficient Disaggregated LLM Inference on Heterogeneous xPUs

Large Language Model (LLM) inference is increasingly served on disaggregated clusters that combine different accelerator types, including NVIDIA GPUs, Huawei Ascend NPUs, and Kunlunxin AI processors. We show that these heterogeneous xPUs exhibit distinct energy-performance behaviors across inference phases and model mo...

DICD-IE: an expert-annotated benchmark for evaluating pretrained biomedical language models in dental implant literature and clinical knowledge extraction

Dental implant literature contains clinically important information on risk factors, treatments, outcomes, and, in rare cases, co-occurrences with oral cancer have been reported. However, this information is typically embedded in concise narrative statements, which complicates automated extraction and limits large-scal...

Visible Consumption: Design, Debugging, and Closed-Loop Remote Control of a Low-Cost IoT Prototype for Industrial Energy and Temperature Monitoring Aligned with SDG 12

Low-cost microcontrollers with integrated wireless connectivity have lowered the barrier to instrumented industrial monitoring, yet production adoption remains constrained by memory, cryptographic capacity, and link reliability. This paper documents the design, implementation, systematic debugging, and closed-loop remo...

Designing and Evaluating Socially Assistive Robots: The Role of Language Choice in Dementia Care Scenarios

Language plays a central role in person-centered care, shaping communication, emotional expression, memory recall, and identity. While socially assistive robots (SAR) show promise for reminiscence-based activities in dementia care, most systems remain monolingual and overlook bilingual users. This work investigated how...

Evaluating Large Language Models for Decision-Making in Agent-Based Urban Mobility Simulations

The "Podcast" ECoG dataset

The "Podcast" electrocorticography (ECoG) dataset captures intracranial neural activity from 9 participants (1,330 grid, depth, and strip electrodes) who listened to a 30-minute naturalistic story containing over 5,000 words. It includes raw recordings, a minimally preprocessed high-gamma version, the auditory stimuli,...

ScholarScribe: A Privacy-First Local LLM Writing Companion for Researchers

ScholarScribe is a desktop application that supports researchers in the preparation of scholarly manuscripts through a suite of local, privacy-preserving tools. The application runs entirely on the user's device, requires no cloud-based application programming interfaces (APIs), collects no telemetry, and transmits no...

CogniGraph: Governed Graph-of-Agents Reasoning over Knowledge Graph Topologies with Semantic SHACL Validation and Constrained F1 Evaluation

Abstract. Multi-agent large language model (LLM) systems are increasingly deployed for regulatoryand high-stakes reasoning, yet none of the prevailing flat-topology frameworks—AutoGen, CrewAI,LangGraph—enforce semantic governance over what each agent may say, in whose name, and withinwhich scope. The result is what we...

Pattern Pressure, Accuracy Drift, and False User-State Attribution

Large-language model outputs are often described as "accurate" or "inaccurate" as if a single axis were being measured. This paper argues that the umbrella word accurate collapses at least four distinct fits: fit to user intent, fit to measured fact, fit to source provenance, and fit to the local conversational pattern...