Optimized data aggregation for a 5M-record dataset using parallel processing on Google Cloud Vertex AI, reducing processing time by 60x through distributed parallelization

Designed a two-step ML pipeline combining classification and regression models to predict clicks-to-impressions ratio, improving baseline by 7x using a custom cost function

Presented results to 50+ peers and evaluated model performance and feature importance to interpret drivers of article engagement, linking predictive insights to potential revenue and content optimization strategies

Developed a convolutional neural network architecture to binarily classify the gaze direction of a subject (looking at camera vs. averted gaze)

Preprocessed a dataset of 5,000+ images using Keras, standardizing resolution and image size to improve model reliability and performance

Tested multiple data augmentation methods such as Gaussian noise, extra dropout layers, and varying image brightness to optimize model precision and reduce overfitting

Built an airport-based MLB travel dataset by mapping teams to IATA airports and computing a full pairwise distance matrix, enabling accurate and consistent travel cost estimation for scheduling

Developed a slot-based mixed-integer optimization model in Gurobi with binary matchup, team location, and travel transition decisions, producing schedules that correctly capture travel across consecutive series

Applied real-world scheduling rules and ran sensitivity benchmarking against observed travel estimates, ensuring feasible itineraries while quantifying the efficiency and fairness tradeoff

Processed 60M+ transactions on a 4-node Spark cluster; exploratory analysis revealed weekday and midnight spikes and 98.5% U.S. volume, informing fraud risk hypotheses

Built a class-weighted Logistic Regression (grid search) achieving 83% recall / 89% accuracy, with LIME highlighting foreign-currency payments (AUD) as top risk driver

Flagged cross-border outliers (USD→JPY) and segmented 7M high-value transactions using Isolation Forest + K-Means, generating tiered monitoring rules for fraud teams

Performed various machine learning methods like Sentiment Analysis to analyze 5 gigabytes of Yelp text data and categorize them into relevant groups

Created Python functions to present users with a relevant review for a selected business based on user preferences

Improved function performance by adjusting text preprocessing parameters, generating more relevant reviews for users from given input

Built production-ready job-market data platform by orchestrating 6 daily Airflow ETL pipelines from the Adzuna API into a BigQuery star schema, delivering reliable, up-to-date analytics for job seekers and recruiters

Implemented semantic resume-to-job matching using vector embeddings, and deployed a K-means clustering pipeline to group jobs by skill similarity using LLM-extracted skills

Shipped a Streamlit app with a RAG hiring assistant that retrieves from live postings to generate data-grounded job descriptions and recommendations, accelerating recruiter workflows

Engineered an interactive five-page Streamlit dashboard that ingests live Yahoo Finance data, lets users assign custom weightings to fundamental vs. technical factors, and returns a normalized 1-10 Investment Rating

Integrated two pre-trained ML pipelines (fundamental and technical) plus a Min-Max scaler to harmonize raw model outputs and enabled re-scoring through a slider-controlled weighting scheme

Unified fragmented Companies House data for 1M+ UK firms using SQL, enabling clear insights into survival, ownership, and industry resilience

Exposed an 80% corporate dissolution rate (2014–2024) and 60% surge in foreign ownership, influencing investor and policy strategy post-Brexit

Developed interactive Tableau dashboards that transformed raw registration data into executive-ready insights, guiding capital allocation across sectors

Utilized Python (pandas) to process, clean, and visualize data for 5 companies from Vanguard S&P 500 ETF (VOO) and 5 from Vanguard Small-Cap Index Fund (VSMAX)

Analysed corporate financial data to explore correlations between KPIs and stock returns, showing that FCF/Sales and EBITDA have weaker correlations with stock price movements

Presented key financial metrics driving market performance using data visualizations, collaborated with team of 6 and facilitated Q&A for audience of 60

Designed and implemented a randomized controlled trial (RCT) to assess how jazz music influences task performance and cognitive metrics

Utilized A/B testing, exploratory data analysis (EDA), regression analysis, and estimated treatment effects (ATE and CATE) with Python

Ran randomized trial (n=62) testing jazz music's effect on focus and accuracy during cognitive tasks