EngD Alumni

The financial industry operates in a highly dynamic and interconnected ecosystem, posing complex challenges for predictive modeling and decision-making. Accurately forecasting financial performance, detecting fraud, assessing credit risk, and ensuring compliance require techniques that can capture temporal, relational, and contextual dependencies in financial data. This dissertation explores the application of Temporal Relational Graph Convolutional Networks (TRGCNs) combined with Financial Knowledge Graphs (FKGs) to address these challenges and advance analytics in finance.

The study introduces FintechKG, a financial knowledge graph constructed through a three-dimensional information extraction pipeline encompassing entities, temporal features, and domain-specific financial relationships. The TRGCN-based framework models both temporal and relational dependencies within FintechKG, enhanced with FinBERT embeddings and social media signals for richer feature representation. Using a financial performance prediction case study, a logistic regression model is applied to classify revenue trends, demonstrating the efficacy of combining relational and textual embeddings.

In addition, the dissertation presents a novel Temporal Credit Knowledge Graph (TCKG) framework to support corporate credit risk assessment. This framework leverages temporal embeddings, relational reasoning, and generative AI techniques to predict credit risk with high accuracy. A comparative analysis involving five models—including logistic regression, Relational Graph Convolutional Networks (RGCNs), generative AI Gemini models, and Graph Recurrent Neural Networks (GRNNs)—highlights the superiority of the Gemini-based approach in terms of performance, robustness, and explainability.

By integrating temporal learning, knowledge graphs, and generative AI, this research contributes a transformative methodology to financial analytics. It sets the stage for future innovations that merge structured reasoning with generative intelligence, enhancing decision-making in complex and dynamic financial environments.

Commentary on Experience in the Programme

The EngD programme at SMU has been an intellectually transformative journey. It allowed me to blend academic rigor with real-world problem-solving across finance and AI. Working under esteemed faculty and collaborating with industry experts helped shape a research methodology that is not only robust but also highly applicable to modern financial systems.

What stood out most was the support structure and interdisciplinary nature of the programme, which encouraged exploration across machine learning, knowledge graphs, and generative AI. The journey culminated in developing a solution with potential real-world impact, and I am proud to carry forward these insights in my work leading AI initiatives at Google Cloud.

I am deeply grateful to the faculty, peers, and administrative team for their guidance and encouragement throughout this journey.

This dissertation looks into the ethical challenges of integrating AI into education, revealing a significant gap in the literature concerning AI's ethical imperatives in educational assessments. It aims to understand the technologies behind assessments, clarify the relationship between AI, ethics, and assessments, and develop a framework for addressing AI's ethical challenges in this setting. The research contributes a detailed examination of AI's role and its ethical consequences in educational assessments, presenting a framework to aid stakeholders to address these complexities. It also calls for further interdisciplinary research and responsible AI application to enhance educational practices ethically and effectively.

The Doctor of Engineering program at SMU is distinguished by its exemplary supervision and robust support systems. My thesis supervisors, Prof. Swapna Gottipati, Prof. Michelle Cheong and Prof. David Lo, provided invaluable guidance, instrumental in both my academic and professional growth. The diverse expertise of the thesis committee members enriched research and professional perspectives, fostering an environment where I felt both challenged and supported both as an academic and industry practitioner.

Additionally, commendations to the diligent logistical and procedural support from the academic administrative staff, in particular Ms. Yeo Lip Pin and Ms. Diana Koh, who were instrumental in streamlining procedural requirements and minimizing administrative hurdles for EngD students. 

I am profoundly grateful for the mentorship and resources that have been pivotal to my education in SMU. This intellectually stimulating programme effectively combines rigorous academic training with comprehensive support, preparing students for impactful careers in their fields.

This study develops a data-driven framework for optimal retail store location planning that integrates road network analysis, mobility data and optimization techniques. By addressing the limitations of traditional approaches that rely on outdated census data and manual site selection, this research offers a scalable and adaptable solution for retail expansion in diverse urban environments.

The research establishes foundational context through comprehensive literature review, identifying significant gaps in community identification methods, store footfall prediction techniques and competitive facility location models. The framework proposes a three-stage spatial partitioning procedure that partitions road networks using the Louvain method, outlines partition boundaries using Uber H3 grids, and classifies partitions through K-means clustering. Experimental results in Da Nang, Vietnam demonstrate the method's effectiveness in organizing large-scale road networks into distinct communities.

The study explores footfall estimation augmentation through mobility data integration, creating population-weighted centroids that enhance the traditional Huff Model commonly used in site selection. This approach harnesses big data's potential, offering a cost-effective and scalable alternative that eliminates reliance on outdated census data and government urban planning records. The methodology employs the Adaptive Large Neighborhood Search (ALNS) algorithm to automate facility allocation across large urban search spaces, achieving near-optimal solutions without exhaustive evaluation of solution spaces.

The integrated framework combines community detection, enhanced Huff Model and ALNS optimization into a cohesive, data-driven approach. This research contributes methodological innovations to academic literature while offering practical solutions to retail expansion challenges, enabling more efficient, accurate and scalable decisions that adapt to diverse urban contexts.

Commentary on Experience in the Programme

The EngD programme at SMU has been a profoundly enriching journey that seamlessly bridged academic rigor with practical industry applications. What distinguished this experience was the exceptional mentorship from my professors, who provided not just technical guidance but also strategic insights that shaped my approach to complex data science challenges. Their close supervision and collaborative spirit created an environment where innovative thinking flourished. The programme's structure encouraged me to tackle real-world problems with academic depth, culminating in a framework that addresses genuine business challenges in retail location planning. This blend of theoretical foundation and practical application has been invaluable in my current role leading data science initiatives at Procter & Gamble.

Government agencies prioritize citizen service delivery to foster trust with the public. Technological advancements, particularly in Artificial Intelligence (AI), hold promise for improving service provision and aligning government operations with citizens' needs. This dissertation contributes a framework for the development of an AI-enabled recommender system known as AI Based Citizen Question-Answer Recommender (ACQAR). Further research was done with the implementation of this system within a Singaporean government agency to enhance the agency’s citizen service delivery. ACQAR integrates Empath X SLA predictor, Citizen Question-Answer system (CQAS), and ChatGPT to generate contextually aware responses for customer service officers. The study aims to optimize government-citizen interactions in the digital age, where citizens expect efficient, personalized, and empathetic services. 

The Doctor of Engineering program (EngD) at Singapore Management University serves as a bridge between industry practices and academic rigor. It fosters and elevates my knowledge and skills in practical implementation within my current work domain, furthering my passion in utilizing Artificial Intelligence and Technology to serve the citizens of Singapore. This program is an ideal choice for those seeking a practical yet academically rigorous learning experience to further their career. 

Deepest gratitude goes to the esteemed supervisory committee members, Vice Provost (Education) Venky Shankararaman, Associate Professor (Education) Ouh Eng Lieh, and Associate Professor Hady Wirawan Lauw. Their boundless patience, unwavering guidance, and steadfast support have been the guiding light of the EngD journey. Professor Michelle Chong and the EngD administration office also deserve heartfelt appreciation for their endless patience and assistance in navigating logistical hurdles and coordinating conference funding arrangements, allowing for full immersion in research endeavors.

The precedence-constrained production scheduling problem (PCPSP) in Long-Term Mine Planning is acknowledged to be NP-hard and conventionally prioritizes the Net Present Value (NPV) of profits. Even so, heightened sustainability concerns necessitate heightened sustainable practices. Yet, research still lags. Hence, we tackled how sustainability elements can be incorporated into the PCPSP, focusing on environmental sustainability of carbon dioxide emissions (or carbon costs).

To begin, our systematic review examined the techniques for the PCPSP for commonalities, trends and sustainability inclusion. We then assessed two Multi-Objective Optimization (MOO) approaches to trade off the NPV of profits against carbon costs -- decomposition-based and domination-based. Under the decomposition-based approach, we used a bounded objective function method and proposed the novel hybrid Temporally Decomposed Greedy Lagrangian Relaxation (TDGLR) algorithm. Next, the domination-based approach compared two popular Multi-Objective Evolutionary Algorithms (MOEAs) of Non-dominated Sorting Genetic Algorithm II and Pareto-Envelope Based Selection Algorithm II using novel heuristics within them. Lastly, we put forth a framework to assess uncertainty within a dual MOEA setup. Overall, our research provides future direction when trading off sustainability elements and incorporating uncertainties for this real-world problem.

The programme was a good bridge between industry and academia whereby research was focused on real-world problems. It instilled the rigour, depth and inquisition that can be applied to and elevate my work as a Data Scientist. The programme also allowed me to connect with other researchers locally and globally so as to expand and exchange ideas.