Projects
Things I've built.
Built an ETL pipeline and interactive Streamlit dashboard for Educational Testing Service, processing 1.7M+ PISA student records across 81 countries and 3 assessment cycles. Implemented two dashboard modes — a guided narrative and a free-explore view — surfacing score distributions and group comparisons rather than misleading averages. Includes a pytest suite with GitHub Actions CI for reproducible validation.
Built a hybrid RAG search assistant over 701K Amazon Beauty reviews and 112K products, fusing BM25 lexical retrieval and FAISS semantic search via Reciprocal Rank Fusion. Llama 3.2 generates grounded recommendations with source attribution linking each result to the reviews and products behind it. Scaled to the full corpus using an IVF index for efficient approximate nearest-neighbour search.
Built and deployed an interactive Shiny for Python dashboard on a Kaggle supply chain dataset from the fashion and beauty industry, exploring inventory patterns and shipping cost drivers across product categories. Built end-to-end from data wrangling through to deployment on Posit Connect.
Data validation Python package published to TestPyPI. Includes a complete CI/CD workflow via GitHub Actions automating unit testing, documentation generation, and publishing pipelines. Published at: test.pypi.org/project/ez_df_data_validator
Real-time computer vision pipeline for an underwater rover to detect submerged golf balls using OpenCV. Achieved sub-second detection latency on live video feed.