Developed a comprehensive macroeconomic model to forecast CPI inflation predictions and recommend optimal interest rate trajectories to the Bank of Canada's Governing Council under diverse macroeconomic scenarios.

Implemented Bayesian Vector Autoregression model with Minnesota priors using 9 macroeconomic variables, incorporating stationarity testing, seasonal adjustment, and posterior sampling with 2,000 draws to quantify forecast uncertainty through fan charts.

Quantified Bank of Canada's policy objectives of stable, low, and predictable inflation into mathematical optimization formula, testing thousands of different policy rate paths via Monte Carlo simulations to identify optimal interest rate trajectory with successful shock simulation capabilities for external economic disruptions.

Demonstrated empirical validation of Solow model predictions showing countries converge toward steady-state equilibrium with temporary deviations during major economic crises (2008/2009, 2020 pandemic).

Analyzed global economic interdependence patterns and regional correlations using a 28-country dataset (1992-2023), examining GDP growth correlations across income groups and identifying varying degrees of economic interdependence between North America, Latin America, and Asia.

Implemented Solow Growth Model framework to analyze capital accumulation dynamics across countries, calculating steady-state capital per worker using production function parameters and examining convergence behavior through law of motion for capital growth.

Investigated the impact of British Columbia's 2008 carbon tax on renewable energy investment levels by analyzing historical wind power development across Canadian provinces, addressing policy questions on the effectiveness of carbon pricing in stimulating green energy investments.

Applied Difference-in-Differences methodology using Canada's Wind Turbine Dataset (2000-2018) to establish causal relationships between carbon taxation and wind energy investment, implementing province fixed effects, three-year lag structures, and multiple control group configurations to address parallel trends assumptions.

Demonstrated that carbon tax implementation generated 552% greater increase in wind-generated electricity in British Columbia relative to control provinces, with stronger long-run effects when accounting for investment-to-output lag, demonstrating that carbon pricing mobilizes private capital toward renewable infrastructure.

Examined Canada's housing affordability crisis by analyzing the impact of immigration on rental market dynamics across 10 provinces, addressing critical policy questions surrounding the role of Canada's significant immigration inflows in exacerbating the housing crisis.

Applied panel regression, Dynamic OLS, and Fully Modified OLS methodologies to provincial data (2000-2023) examining causal relationships between immigration inflows and rental housing costs, incorporating fixed effects and autocorrelation corrections to mitigate endogeneity concerns.

Demonstrated that 1% population increase from immigration generates 0.236% rental price increase, with pronounced long-run versus short-run effects, establishing immigration as statistically significant yet moderate factor in housing cost escalation.

Finance team needed to query databases for payments and reconciliation but lacked SQL knowledge, creating dependencies on IT department for routine data requests.

Built natural language to SQL conversion system using LlamaIndex RAG architecture with vector embeddings of database schema and historical query examples for accurate query generation.

Implemented LLaMA 3 model with custom fine-tuning on city's financial database structure and business rules, enabling context-aware SQL generation with validation and safety guardrails.

Created automated learning system that improves query accuracy over time by analyzing user feedback and maintaining conversation history for iterative query refinement.

Analysts manually review thousands of water accounts quarterly to identify potential leaks or meter failures, creating time-intensive workload with risk of missing critical issues.

Built hybrid machine learning system combining supervised Random Forest and Gradient Boosting models trained on historical leak data with unsupervised Isolation Forest algorithms for comprehensive anomaly detection.

Implemented feature engineering processing consumption patterns, seasonal variations, and property characteristics with ensemble methods combining multiple algorithm outputs for improved accuracy.

Applied intelligent filtering using neighborhood consumption averages and account-specific rules to reduce false positives by 70%, generating automated reports with confidence scores.

Manual processing of property exemption forms creates backlogs and inconsistent data entry, requiring time-intensive cross-referencing with city records.

Built automated form processing system using Azure Computer Vision API for OCR text extraction with custom validation rules for address formatting and account verification.

Implemented fuzzy string matching using rapidfuzz library to correlate extracted owner and property data with city records, handling variations in formatting with confidence scoring.

Generated standardized summaries with verification status and discrepancy flagging, eliminating processing backlogs and improving data consistency.

Finance team manually processes discount applications with inconsistent data entry, causing verification delays and approval workflow bottlenecks.

Implemented Azure Computer Vision API with custom preprocessing to handle various form layouts and handwriting recognition for improved OCR accuracy.

Applied pattern recognition using regex validation and confidence scoring to extract applicant names, addresses, and account numbers with error detection capabilities.

Built automated validation system cross-referencing extracted data against city property records, flagging discrepancies and reducing manual review time by 80%.

Utilities team manually inputs account and meter serial numbers from meter inventory forms, creating time-consuming and error-prone data entry processes that delay billing updates.

Built automated processing system using Azure Computer Vision API achieving 99%+ accuracy on structured utility forms with intelligent data extraction capabilities.

Implemented regex pattern matching and rapidfuzz fuzzy string matching algorithms to reliably capture account numbers and serial numbers despite OCR variations.

Created automated output system generating formatted Excel spreadsheets and organized PDF files for seamless distribution to billing teams.