(Bruce) Phuong Quoc Anh To

• Reference - Dr. Mehdi Hosseinzadeh [Capstone Project Supervisor]

• Reference - Assoc. Prof. Cao Tuan Dung [Capstone Project Supervisor]

• Architected and developed an AI-powered Knowledge Assistant for asset finance brokers using a RAG pipeline, reducing lender policy query resolution from 15-30 minutes to under 30 seconds; project reached MVP-ready status at departure.
• Delivered full-stack Salesforce (AFOS) development across CRM workflows, spanning automated broker notifications, duplicate lead detection, DOF calculation logic, and a custom bulk record transfer tool.
• Tools/Skills: LLM, RAG, LangChain, Vector Databases, FastAPI, Python, Prompt Engineering, Word Embeddings, SOQL, Apex, Lightning Web Components, Agile/Scrum.

• Reference - Bill Tsouvalas [Managing Director]

• Addressed technical debt by testing and validating OCR APIs using Postman, restructuring project source code for improved maintainability, and managing application lifecycle including port management and status persistence.
• Developed significant frontend and backend features including profile management, logout flows, image/audio detail pages, and database CRUD operations.
• Tools/Skills: Next.js, Django, Python, Optical Character Recognition (OCR), RESTful APIs.

• Reference - Ankit Yadav [CTO]

• Modernised and enhanced the electricity Customer Management Information System (CMIS) 4.0 serving nearly 30M customers nationwide, including households and enterprises, across 1,200+ operational units in Vietnam.
• Led a team of 6 engineers to deploy mobile solutions across 5+ provinces in northern Vietnam as technical lead for the CMIS Mobile project.
• Maintained and enhanced the EVN Gateway system, ensuring secure API interactions for customer authentication and data access with the National Population Database under Vietnam's Ministry of Public Security.
• Delivered high-performing software solutions scaling to handle 1M+ service requests annually, ensuring compliance with national data security standards.
• Tools/Skills: Angular, React Native, TypeScript, Java EE 8, ASP.NET, SQL, Oracle Database, MongoDB Realm, WebLogic Server, Git, Jenkins, SVN, Microservices, Azure DevOps, SDLC, Privileged Identity Management.

• Collaborated with a team of 7 engineers to develop an online Learning Management System (Eclazz), serving 1M+ learning materials for over 10K students and trusted by 1,340+ teachers across 320+ organisations.
• Designed and implemented a Content Management System for a gaming portal, improving content delivery workflows.
• Tools/Skills: HTML, CSS, Bootstrap, JavaScript, ReactJs, Redux, Node.js, Express.js, RESTful APIs, MongoDB, CI/CD, Docker, Kubernetes.

Object-Based Visual SLAM for Urban Tram Navigation

• Developed an end-to-end Visual SLAM pipeline integrating YOLOv8 object detection, ByteTrack multi-object tracking, and CubeSLAM-based 3D cuboid reconstruction into ORB-SLAM3, achieving 0.52 mean 3D IoU and 65% precision on KITTI odometry sequences.
• Demonstrated cross-dataset generalisation by training on BDD100K (100K images) and evaluating on KITTI; ByteTrack integration reduced redundant map entities by 81.2% through persistent object identity tracking.
• Containerised the ORB-SLAM3 development environment using Docker with X11 forwarding, enabling real-time 3D visualisation and deployment on Adelaide urban footage.

Hybrid Recommender System for Grocery Retail

• Built a hybrid recommender combining FP-Growth frequent pattern mining and item-based collaborative filtering (KNN, cosine similarity) with recency-aware personalisation on around 38,600 grocery transactions.
• Hybrid model outperformed standalone collaborative filtering across key metrics: Precision@5 (0.2040 vs. 0.2014), Recall@5 (0.1911 vs. 0.1886), and Hit Rate@5 (0.6529 vs. 0.6476), while addressing cold start and 98.86% matrix sparsity via frequent-itemset fallback and eager pre-computation for real-time inference.

RNNs for Stock Price Prediction

• Designed and optimised a Recurrent Neural Network (RNN) model using Stacked LSTM architectures in Python with TensorFlow and Keras, achieving a 47% reduction in MSE for stock price forecasting vs. traditional methods.

Diabetes Prediction using Perceptron

• Optimised a Perceptron algorithm for diabetes prediction, achieving 74.68% accuracy on the Pima Indians Diabetes dataset through hyperparameter tuning, demonstrating mastery of supervised learning fundamentals.