An SQL query engine built from scratch in Python (lexer, parser, query planner and execution pipeline) that runs real queries against CSV files. It supports filtering, sorting, arithmetic expressions, aliases, and grouped aggregates with every feature verified for correctness against SQLite.
An interactive Tableau dashboard exploring maritime activity and emissions across Poland's four seaports. Working from Statistics Poland's experimental AIS dataset, I engineered the underlying metrics in a Jupyter notebook before visualising port specialisation, trade flows, seasonal rhythms, and emissions across Gdańsk, Gdynia, Szczecin, and Świnoujście.
An NBA tool that finds the most flattering yet technically true stat line in any player's career. It beam-searches a subset of a player's Pareto efficient box scores, pins each stat threshold to his own numbers then stacks constraints until almost no one else in NBA history did the same. It was computed across 1.6 million player-games from 1947 to today and precomputed in parallel for instant lookup.
A movie "anti" recommendation system that finds films with opposite genre characteristics using an autoencoder. The system trains a PyTorch autoencoder on MovieLens data (weighting genres by inverse document frequency so rare, distinctive genres carry more signal than ubiquitous ones) to learn a compressed latent representation, then ranks films by lowest cosine similarity in that embedding space to surface genre opposites.
The best way to reach me is through my email:
otisfraser@protonmail.com