Featured Projects
AI-driven solutions and data science projects that showcase my expertise in machine learning, finance, and software development.
Alopecia Areata Risk Model & Shiny App
Bioinformatics pipeline analyzing gene expression data from GSE68801 dataset. Features LASSO feature selection, multiple ML models (Random Forest, SVM, GLMNET), and interactive Shiny web application for Alopecia Areata risk assessment.
Problem
Alopecia Areata affects 2% of the population, but current diagnostic methods are limited and subjective. Need for early detection using molecular biomarkers from gene expression data.
Approach
Applied differential expression analysis, LASSO feature selection, and multiple ML algorithms to gene expression data. Deployed interactive Shiny web application for real-time risk assessment.
Impact
Achieved 92% AUC with 85% accuracy, identifying 21 key genes associated with AA risk. Deployed Shiny app enables real-time risk assessment for early diagnosis.
Exam Portal System
A comprehensive online examination platform built during COVID-19 school closures to enable secure remote assessments. Features advanced anti-cheat detection, beautiful responsive UI, and dynamic dashboards for students and faculty.
Problem
During COVID-19 school closures, educational institutions needed a secure, reliable online examination platform to conduct assessments remotely while preventing cheating and ensuring academic integrity
Approach
Developed a comprehensive full-stack web application with Django backend, responsive frontend, advanced anti-cheat detection, and real-time monitoring to enable secure remote examinations
Impact
Enabled educational institutions to conduct secure online exams during COVID-19 lockdowns, serving thousands of students with anti-cheat detection, rate limiting, and comprehensive user management
Stock Volatility Prediction
Engineered features and trained Linear Regression, Random Forest, XGBoost models. Added GARCH(1,1) for time-varying volatility analysis.
Problem
Predicting stock price volatility for better risk management and trading decisions
Approach
Feature engineering from market data, ensemble ML models, and GARCH time series analysis for volatility clustering
Impact
Improved volatility prediction accuracy by 23% compared to baseline models, enabling better risk assessment
Capital Structure Analysis - CSL Limited
CSL Limited (ASX: CSL) 2019-2024 peer comparison with leverage drivers analysis and strategic recommendations.
Problem
Understanding CSL's capital structure and providing strategic recommendations for optimal financing
Approach
Comprehensive analysis of leverage ratios, peer comparison, and identification of key capital structure drivers
Impact
Provided actionable recommendations that could optimize CSL's cost of capital by 15-20 basis points
SlideSmith - Multi-Agent AI Slide Maker
A production-ready, distributed multi-agent system for automated slide deck generation with advanced quality assurance, semantic validation, and multi-format export. Built on modular, extensible architecture supporting both cloud and edge LLM deployments with 13 specialized AI agents.
Problem
Enterprise teams struggle with presentation creation requiring hours of work, inconsistent quality, lack of automated validation, and inability to scale. Existing tools lack fact-checking, accessibility compliance, and flexible deployment options for privacy-sensitive environments.
Approach
Implemented a distributed 13-agent collaborative pipeline using LLM orchestration patterns with intelligent model routing (Phi-4 14B for research, Gemma3 4B for content). Features DAG-based workflow, parallel QA validation (4 concurrent agents), provider abstraction (Ollama/OpenAI), and advanced PPTX engine with native chart rendering and smart text wrapping.
Impact
Delivers enterprise-grade presentation generation with 75% faster parallel QA, 60% cost optimization through smart routing, 99.5% reliability with graceful degradation, and 100% privacy with local-first deployment. Automated fact-checking, WCAG 2.1 compliance, and Flesch-Kincaid readability analysis ensure production-ready output in 3-5 minutes per deck.
Wizard Tower Defense
A strategic tower defense game built with Java and Processing, featuring wave-based combat, progressive difficulty, and mana economy system. Defend the wizard's house from waves of monsters (gremlins, beetles, worms) by strategically placing and upgrading towers along their path.
Problem
Traditional tower defense games lack depth in strategic decision-making and often have limited upgrade paths. Need for a game that combines resource management, strategic tower placement, and progressive difficulty with clean architecture and extensibility.
Approach
Developed a component-based architecture using Processing framework with layered rendering system (7 layers), controller pattern for game logic, and JSON-driven configuration. Implemented mana economy system, multiple upgrade paths per tower (range/speed/damage), and wave-based progression with 3 enemy types.
Impact
Created an engaging tower defense game with strategic depth through individual tower upgrades, resource management via mana economy, and progressive difficulty across 4 custom-designed levels. Clean MVC architecture enables easy extension with new monsters, towers, and mechanics.
Additional Projects
EDU Chat - Student Collaboration App
Student collaboration application with full project management artifacts including business case, WBS, risk assessment, and quality planning.
Drone Content & UAV Operations
Content strategy and compliance research for Australian UAV regulations. Social media operations and regulatory compliance management.