This is to certify that the work presented in this thesis, titled, "Evaluating the Multilingual Capabilities of LLM-based Web UI Agents", is the outcome of the investigation and research carried out by me. It is also declared that neither this thesis nor any part thereof has been submitted anywhere else for the award of any degree, diploma or other qualifications.

Evaluating the Multilingual Capabilities of LLM-based Web UI Agents

The proliferation of misinformation and disinformation, commonly referred to as fake news, poses a significant threat to democratic processes and public trust. Automated and accurate detection of such content is a critical challenge in Natural Language Processing (NLP). Traditional machine learning (ML) models often struggle to capture the complex semantic and contextual nuances inherent in deceptive language. This paper addresses this limitation by proposing and evaluating a robust fake news detection framework utilizing advanced transformer-based models: Bidirectional Encoder Representations from Transformers (BERT) and its resource-efficient variant, DistilBERT. We fine-tune these models on established benchmark datasets, such as LIAR, and compare their performance against conventional ML baselines, including Support Vector Machines (SVM) and Naïve Bayes. Our methodology incorporates the intricacies of WordPiece tokenization and the power of the multi-head self-attention mechanism to generate highly contextualized text representations. The experimental results demonstrate that both BERT and DistilBERT significantly outperform the baselines across key metrics-Accuracy, Precision, Recall, and Fscore-with DistilBERT achieving comparable performance to the full BERT model while offering substantial computational efficiency. This research contributes a detailed comparative analysis and a high-performing, deployable solution for mitigating the spread of online misinformation.

A Contextualized Approach to Fake News Detection: Comparative Analysis of BERT

Climate change is increasingly disrupting rural livelihoods and reinforcing existing gender inequalities, particularly in vulnerable regions. This study investigates the gender-specific impacts of climate change on rural communities in Muzaffargarh District, Pakistan-an area frequently exposed to extreme weather events such as floods, droughts, and heatwaves. Using a mixed-methods approach, data were collected from 110 respondents (55 men and 55 women) through structured questionnaires, focus group discussions, and key informant interviews. Findings reveal significant gender disparities in land ownership, decision-making power, CSA awareness, and institutional access. Only 18% of women owned agricultural land, compared to 82% of men, limiting their capacity to access credit and adopt CSA practices. Women also reported lower confidence in adapting to climate change and relied more on informal coping strategies, such as unpaid labor and community support, while men tended to migrate or mechanize operations. Despite these challenges, 76% of respondents expressed willingness to adopt CSA practices if adequate financial or technical support were provided. However, preferred support mechanisms varied: women prioritized training and input provision, while men favored mechanization and market access. The study also found that institutional support was insufficient, especially for women, due to limited outreach by extension services and a lack of gender-sensitive programming. This research underscores the need for climate adaptation policies that are both gender-inclusive and context-specific. It recommends targeted interventions such as female-focused extension services, accessible financial tools, and genderdisaggregated planning to enhance equitable climate resilience in rural Pakistan.

Climate Change and Gender Roles in Rural Communities A study in Muzaffargarh

In partial fulfillment of the requirements for the award of the Master of Arts Degree in History.

MATRIARCHAL DOMINANCE IN THE KINGDOMS OF KOMBO AND NIUMI IN PRE-COLONIAL SENEGAMBIA

Probability is a cornerstone of biostatistics, providing a mathematical framework to quantify uncertainty in clinical practice. This paper highlights the importance of probability in clinical medicine, focusing on its role in diagnosis, treatment, and prognosis. Examples from medical testing and therapeutic decision-making illustrate how probabilistic reasoning enhances accuracy, reduces errors, and supports evidencebased medicine. The paper concludes with recommendations for integrating probabilistic training into medical curricula to improve healthcare quality.

The Importance of Using Probabilities in Clinical Medicine: A 
Biostatistical Perspective for Evidence-Based Decision Making

In Mumbai's fast-growing private real estate construction sector, persistent challenges-cost overruns, schedule delays, and inconsistent quality-continue to limit project performance despite rising demand and increasing urban pressures. Artificial Intelligence (AI) has emerged globally as a transformative tool capable of reshaping construction planning, execution, and monitoring. Yet, in Mumbai, AI adoption remains at a formative stage, shaped by a complex interplay of technological limitations, cultural resistance, and organisational readiness. This study explores how AI is currently being used, where it creates value, and what barriers must be overcome for meaningful transformation.

A mixed-methods research design was employed to capture both the breadth and depth of AI adoption. Quantitative insights were gathered through a structured survey of 99 construction professionals, spanning developers, engineers, consultants, and project managers. To complement this, qualitative interviews and focus group discussions were conducted with industry experts to understand their lived experiences, perceptions, and concerns regarding AI-enabled practices. Data were analysed using descriptive statistics, factor analysis, and thematic coding to produce an integrated, evidence-based understanding of AI's real-world impact within Mumbai's construction environment.

Findings reveal that while AI adoption is still emerging, its footprint is steadily expanding.

The most recognised and frequently applied AI tools include predictive analytics for cost estimation, automated scheduling systems, and computer-vision-based quality inspections.

Respondents involved in AI-enabled projects reported heightened confidence in the technology's potential to enhance efficiency, reduce rework, and improve decision-making.

However, this optimism exists alongside significant obstacles. The study identifies notable barriers such as low digital literacy, fragmented data systems, regulatory ambiguity, and organisational cultural resistance. Many firms struggle to integrate AI into legacy workflows, and small and medium-sized enterprises face higher financial and technical hurdles. The discussion highlights that successful AI-enabled transformation requires more than just technological investment-it demands structural, cultural, and behavioural shifts within organisations. AI's impact is therefore as socio-technical as it is operational, requiring alignment across people, processes, and platforms. This research confirms that AI holds strong promise for reducing chronic inefficiencies in Mumbai's real estate construction sector. Yet, the gap between theoretical potential and onground performance remains wide. To bridge this divide, organisations must adopt a phased, context-appropriate strategy that prioritises digital literacy, data standardisation, regulatory clarity, and targeted workforce upskilling. The study offers a practical implementation roadmap tailored to Mumbai's unique ecosystem, serving as a valuable resource for developers, project managers, policymakers, and technology providers. Ultimately, AI is positioned not as a replacement for human expertise, but as a powerful enabler of smarter, safer, and more resilient urban development.

ASSESSING THE CAPABILITIES OF AI IN PRIVATE REAL ESTATE DEVELOPMENT WITHIN THE CONSTRUCTION SECTOR

In today’s digital era, the rapid spread of misinformation poses a significant challenge, influencing public perception and decision-making. This project focuses on developing an automated Fake News Detection System using machine learning techniques to differentiate between genuine and misleading news articles. The system follows a structured pipeline, beginning with data preprocessing, followed by feature extraction, and finally, classification using multiple machine learning algorithms.
The dataset used consists of labeled news articles, categorized as either real or fake. Preprocessing techniques such as text normalization, removal of stop words, punctuation elimination, and lemmatization are applied to enhance data quality. To convert textual data into numerical representations, Term Frequency-Inverse Document Frequency (TF-IDF) is utilized, ensuring that significant terms contribute effectively to classification. Several machine learning models, including Logistic Regression, Passive Aggressive Classifier, Multinomial Naïve Bayes, and Support Vector Machine (SVM), are implemented to assess their effectiveness in detecting fake news.
Among these models, the Passive Aggressive Classifier demonstrates the highest accuracy, making it a suitable choice for real-time fake news identification. Performance evaluation is conducted using metrics such as accuracy, precision, recall, and confusion matrix analysis to determine the efficiency of each model.
The developed system offers a reliable solution to counter misinformation, with potential applications in journalism, social media platforms, and news verification tools. Future enhancements could involve deep learning models, real-time web scraping, and multilingual support to further improve detection capabilities. This project serves as a crucial step toward mitigating the negative impact of false information in the digital landscape.
