Agentic Ecommerce AI — Bug Resolution & Production Delivery

Python · Regex · BeautifulSoup4 · Git · GitHub

2 PRs merged 0 regressions 6 retailers

Diagnosed and fixed 3 critical routing and ranking defects in a live AI shopping agent serving 6 Australian retailers. Restructured 12-branch conditional logic, implemented regex-based model number extraction covering iPhone, Galaxy, and Pixel variants, and added a relevance filter (threshold: 0.05) with fallback safeguard. 2 pull requests merged into main — fix confirmed operational 10 hours post-merge under live production conditions.

Python Regex BeautifulSoup4 Root Cause Analysis GitHub

US Consumer Expenditure Forecasting — Financial Trend Analysis

R · ARIMA/ETS · SARIMA · Power BI

MAPE <0.2% Top quartile in cohort

Evaluated 3 time series models (ARIMA, ETS, SARIMA) using AIC/BIC model selection in R. Conducted residual diagnostics, Ljung-Box testing, and 6-period out-of-sample backtesting across 12 product categories. Achieved MAPE of 0.18% on held-out test data. Produced a 12-page client-ready report with 8 graphical outputs and scenario commentary.

Forecasting Time Series R Power BI

Tesbury: Retail Strategy & Investment Analysis

R · MCDA (WSM/TOPSIS) · OLS Regression

8 sites ranked R² = 0.74 Board-level delivery

Applied WSM and TOPSIS to rank 8 potential site investment options across 6 weighted criteria. Built OLS regression model identifying 4 statistically significant sales drivers (R² = 0.74). Conducted sensitivity analysis varying top-3 weights by ±20% to stress-test recommendations. Delivered ranked investment recommendations with supporting financial business case to a 5-person senior panel.

Consulting MCDA Regression Options Analysis

Counterfeit Review Detection — NLP Classification Pipeline

Python · SVM · TF-IDF · Scikit-learn

94% accuracy 10,000+ records End-to-end pipeline

Preprocessed 10,000+ unstructured product reviews — tokenisation, stop-word removal, lemmatisation. Applied TF-IDF vectorisation across a 5,000-feature space, trained SVM classifier with RBF kernel and 5-fold cross-validation grid search. Achieved 94% accuracy (Precision: 0.93, Recall: 0.95, F1: 0.94). Produced governance documentation covering methodology, assumptions, and responsible use.

NLP Machine Learning Python Classification

Technology Adoption in Real Estate — Qualitative & Sentiment Research

Python (NLTK, TextBlob) · Thematic Analysis

20 stakeholder interviews Coherence 0.68

Designed and executed a 20-interview primary research programme investigating technology adoption behaviour in real estate. Applied NLP sentiment analysis (NLTK) and thematic coding to identify adoption barriers and enablers. Achieved topic coherence score of 0.68. Identified 5 key adoption barriers and delivered strategic recommendations as a structured consultancy report.

Qualitative Research Thematic Analysis Sentiment Analysis Stakeholder Management